# ModelScale

> Source-aware model, pricing and workload evidence for practical AI decisions, published by MonoMind AI Lab.

ModelScale reproduces what upstream sources publish about large language models. Nothing here is modelled, estimated or interpolated. Each model carries 7 capability axes (Agentic, Coding, Knowledge, Reasoning, Multimodal & Grounded, Instruction Following, Math), list token prices per million tokens, measured time to first token and output throughput, a context window and a release date. The upstream projects are BenchLM, LiveBench and OpenRouter; every fetch is dated, and the date is shown beside the value rather than behind it.

Each value carries an evidence status. `observed` means a source published it on the date shown. `retained` means the last good reading is being shown because the newest fetch did not carry that field. `unavailable` means no source publishes it, and `failed` means the fetch itself did not succeed. The last two render as a missing value — never a zero, never an invented number. Where a line in this file omits a figure, that is why.

## Pages

- [ModelScale — evidence for AI runtime and cost decisions](https://modelscale.dev/): Compare AI model capability, runtime and price against the sources each figure came from. A measurement nobody published stays unavailable, never a zero.
- [Models workbench](https://modelscale.dev/models): Filter the model catalog by provider, access and capability, read the price-performance frontier, and carry two to four models into a shareable comparison.
- [Popular models leaderboard](https://modelscale.dev/popular-models): Rank models across seven capability axes with sortable columns, per-model benchmark rows and quality-against-cost plots. Export the table as CSV or PNG.
- [Compare models](https://modelscale.dev/compare): Put two to four models side by side: capability scores, runtime measurements, every published price component, and the pair deltas the source itself computed.
- [Make it yours](https://modelscale.dev/make-it-yours): Weight the seven capability axes and throughput, set runtime thresholds, and rank every model by the weighted sum of the scores its sources actually measured.
- [Subscribe vs API](https://modelscale.dev/subscribe-vs-api): Price a monthly workload against published subscription plans and API token rates at once, with itemised lines and the breakeven volume where the two cross.
- [Model lifecycle radar](https://modelscale.dev/model-lifecycle): Announced and observed model retirements with the source that published each date, the replacement each provider named, and how long is left before it lands.
- [Articles](https://modelscale.dev/articles): Guides and analysis on model routing, usage monitoring and cost control, written against the same evidence rules as every figure elsewhere on the site.
- [Cited sources](https://modelscale.dev/sources): Every upstream fetch behind the site: what each source supplies, when it was last read, how many records it returned, and the rules that govern its values.

## API

One JSON endpoint is public on this domain:

- [GET /api/model/{id}](https://modelscale.dev/api/model/claude-fable-5-1): One model's full evidence record, the same shape the profile page renders.

The catalog API behind every page is reached over an internal service binding and has no published address on this domain, so its read paths are named here rather than linked:

- `GET /api/v1/catalog` — Every model with its full evidence record: capability scores, prices, runtime measurements, and the source and date behind each value.
- `GET /api/v1/catalog?view=summary` — The same catalog projected down to what a list needs. Every page that renders many models at once reads this rather than the full record, and it is the one to prefer unless you need per-benchmark detail.
- `GET /api/v1/models/{id}` — One model's profile: benchmark task rows, source links, and lifecycle history.
- `GET /api/v1/compare?models={ids}` — Pairwise category deltas for two to four comma-separated model ids, as the source published them rather than derived here.
- `GET /api/v1/plans` — Published subscription plans with their charge, cadence, and the page each price was read from.
- `GET /api/v1/lifecycle` — Release and retirement notices with announced and effective dates, and a replacement id where the provider named one.
- `GET /api/v1/articles` — The editorial records behind /articles, as structured sections and blocks.

## Models

- [Claude Fable 5.1 (Anthropic) — 84.76](https://modelscale.dev/model-profile?model=claude-fable-5-1)
- [GPT-6 Astra (OpenAI) — 82.94](https://modelscale.dev/model-profile?model=gpt-6-astra)
- [Claude Opus 5 (Anthropic) — 81.83](https://modelscale.dev/model-profile?model=claude-opus-5)
- [Claude Fable 5 (Anthropic) — 81.42](https://modelscale.dev/model-profile?model=claude-fable-5)
- [GPT-5.6 Sol (OpenAI) — 80.7](https://modelscale.dev/model-profile?model=gpt-5-6-sol)
- [Gemini 3.8 Flash (Google) — 75.48](https://modelscale.dev/model-profile?model=gemini-3-8-flash)
- [Kimi K3 (Moonshot AI) — 74.63](https://modelscale.dev/model-profile?model=kimi-3)
- [Claude Opus 4.8 (Anthropic) — 72.34](https://modelscale.dev/model-profile?model=claude-opus-4-8)
- [GPT-5.5 (OpenAI) — 72.09](https://modelscale.dev/model-profile?model=gpt-5-5)
- [Qwen3.8 Max (Alibaba) — 71.76](https://modelscale.dev/model-profile?model=qwen3-8-max)
- [GPT-5.6 Terra (OpenAI) — 71.13](https://modelscale.dev/model-profile?model=gpt-5-6-terra)
- [GPT-5.4 (OpenAI) — 71](https://modelscale.dev/model-profile?model=gpt-5-4)
- [Claude Opus 4.7 (Anthropic) — 70.6](https://modelscale.dev/model-profile?model=claude-opus-4-7)
- [Muse Spark 1.2 (Meta) — 70.41](https://modelscale.dev/model-profile?model=muse-spark-1-2)
- [Muse Spark 1.1 (Meta) — 70.31](https://modelscale.dev/model-profile?model=muse-spark-1-1)
- [Gemini 3.1 Pro (Google) — 70.17](https://modelscale.dev/model-profile?model=gemini-3-1-pro)
- [Grok 4.6 (xAI) — 69.9](https://modelscale.dev/model-profile?model=grok-4-6)
- [Claude Sonnet 5 (Anthropic) — 69.88](https://modelscale.dev/model-profile?model=claude-sonnet-5)
- [Claude Opus 4.7 (Adaptive) (Anthropic) — 69.39](https://modelscale.dev/model-profile?model=claude-opus-4-7-max)
- [Claude Opus 4.6 (Anthropic) — 69.29](https://modelscale.dev/model-profile?model=claude-opus-4-6)
- [GPT-5.2 Pro (OpenAI) — 68.98](https://modelscale.dev/model-profile?model=gpt-5-2-pro)
- [Gemini 3.7 Flash (Google) — 68.64](https://modelscale.dev/model-profile?model=gemini-3-7-flash)
- [Gemini 3.6 Flash (Google) — 68.54](https://modelscale.dev/model-profile?model=gemini-3-6-flash)
- [Gemini 3.5 Flash (Google) — 68.16](https://modelscale.dev/model-profile?model=gemini-3-5-flash)
- [Grok 4.5 (xAI) — 68.03](https://modelscale.dev/model-profile?model=grok-4-5)
- [Muse Spark (Meta) — 67.65](https://modelscale.dev/model-profile?model=muse-spark)
- [Grok 4.20 (xAI) — 67.27](https://modelscale.dev/model-profile?model=grok-4-20-beta)
- [GLM-5.3 (Z.AI) — 67.07](https://modelscale.dev/model-profile?model=glm-5-3)
- [Qwen3.7 Max (Alibaba) — 67.03](https://modelscale.dev/model-profile?model=qwen3-7-max)
- [GLM-5.2 (Z.AI) — 66.82](https://modelscale.dev/model-profile?model=glm-5-2)
- [DeepSeek V4 Pro 0813 (DeepSeek) — 66.57](https://modelscale.dev/model-profile?model=deepseek-v4-pro-max)
- [Gemini 3 Pro (Google) — 66.49](https://modelscale.dev/model-profile?model=gemini-3-pro)
- [GLM-5.3-Flash (Z.AI) — 66.04](https://modelscale.dev/model-profile?model=glm-5-3-flash)
- [Kimi K2.7 Code (Moonshot AI) — 65.61](https://modelscale.dev/model-profile?model=kimi-k2-7-code)
- [Kimi K2.6 (Moonshot AI) — 65.51](https://modelscale.dev/model-profile?model=kimi-2-6)
- [Ornith-1.5-397B (Ornith AI) — 65.13](https://modelscale.dev/model-profile?model=ornith-1-5-397b)
- [GPT-5.2 (OpenAI) — 64.97](https://modelscale.dev/model-profile?model=gpt-5-2)
- [GPT-5.6 Luna (OpenAI) — 64.66](https://modelscale.dev/model-profile?model=gpt-5-6-luna)
- [Qwen3.8-27B (Alibaba) — 64.65](https://modelscale.dev/model-profile?model=qwen3-8-27b)
- [GPT-5.3 Codex (OpenAI) — 64.19](https://modelscale.dev/model-profile?model=gpt-5-3-codex)
- [dots3-note Preview (Dots Studio) — 64.09](https://modelscale.dev/model-profile?model=dots3-note-preview)
- [Qwen 3.6 Max (preview) (Alibaba) — 63.83](https://modelscale.dev/model-profile?model=qwen3-6-max-preview)
- [GLM-5.1 (Z.AI) — 63.35](https://modelscale.dev/model-profile?model=glm-5-1)
- [GPT-5.1 (OpenAI) — 63.24](https://modelscale.dev/model-profile?model=gpt-5-1)
- [Claude Sonnet 4.6 (Anthropic) — 62.95](https://modelscale.dev/model-profile?model=claude-sonnet-4-6)
- [MiMo-V2.5-Pro (Xiaomi) — 62.89](https://modelscale.dev/model-profile?model=mimo-v2-5-pro)
- [MiMo-V2-Pro (Xiaomi) — 62.79](https://modelscale.dev/model-profile?model=mimo-v2-pro)
- [Claude Opus 4.6 (Adaptive) (Anthropic) — 62.01](https://modelscale.dev/model-profile?model=claude-opus-4-6-thinking)
- [Gemini 3 Flash (Google) — 62](https://modelscale.dev/model-profile?model=gemini-3-flash)
- [Qwen3.7 Plus (Alibaba) — 61.78](https://modelscale.dev/model-profile?model=qwen3-7-plus)
- [GLM-5-Turbo (Z.AI) — 61.69](https://modelscale.dev/model-profile?model=glm-5-turbo)
- [GPT-5.5 Pro (OpenAI) — 61.68](https://modelscale.dev/model-profile?model=gpt-5-5-pro)
- [GLM-5 (Z.AI) — 61.53](https://modelscale.dev/model-profile?model=glm-5)
- [MiniMax M3 (MiniMax) — 61.47](https://modelscale.dev/model-profile?model=minimax-m3)
- [Hy4 preview (Tencent) — 61.16](https://modelscale.dev/model-profile?model=hy4-preview)
- [GPT-5.4 mini (OpenAI) — 61.13](https://modelscale.dev/model-profile?model=gpt-5-4-mini)
- [Hy3 (Tencent) — 61.06](https://modelscale.dev/model-profile?model=hy3)
- [Inkling (Thinking Machines Lab) — 60.52](https://modelscale.dev/model-profile?model=inkling)
- [Qwen3.6 Plus (Alibaba) — 60.48](https://modelscale.dev/model-profile?model=qwen3-6-plus)
- [GPT-5.4 Pro (OpenAI) — 60.27](https://modelscale.dev/model-profile?model=gpt-5-4-pro)
- [Grok 4.3 (xAI) — 59.81](https://modelscale.dev/model-profile?model=grok-4-3)
- [GPT-5.4 nano (OpenAI) — 59.52](https://modelscale.dev/model-profile?model=gpt-5-4-nano)
- [Gemini 3 Pro Deep Think (Google) — 59.1](https://modelscale.dev/model-profile?model=gemini-3-pro-deep-think)
- [Inkling-Small (Thinking Machines Lab) — 58.95](https://modelscale.dev/model-profile?model=inkling-small)
- [Gemini 3.5 Flash-Lite (Google) — 58.84](https://modelscale.dev/model-profile?model=gemini-3-5-flash-lite)
- [Claude Opus 4.5 (Anthropic) — 58.55](https://modelscale.dev/model-profile?model=claude-opus-4-5)
- [MiMo-V2.5 (Xiaomi) — 57.88](https://modelscale.dev/model-profile?model=mimo-v2-5)
- [GLM-4.7 (Z.AI) — 57.82](https://modelscale.dev/model-profile?model=glm-4-7)
- [MiMo-V2-Omni (Xiaomi) — 57.78](https://modelscale.dev/model-profile?model=mimo-v2-omni)
- [Grok 4.1 (xAI) — 57.76](https://modelscale.dev/model-profile?model=grok-4-1)
- [GLM-5V-Turbo (Z.AI) — 57.66](https://modelscale.dev/model-profile?model=glm-5v-turbo)
- [GLM-5 (Reasoning) (Z.AI) — 57.59](https://modelscale.dev/model-profile?model=glm-5-reasoning)
- [Gemini 2.5 Pro (Google) — 57.37](https://modelscale.dev/model-profile?model=gemini-2-5-pro)
- [Qwen3.5 397B (Reasoning) (Alibaba) — 57.31](https://modelscale.dev/model-profile?model=qwen3-5-397b-reasoning)
- [Kimi K2.5 (Reasoning) (Moonshot AI) — 57.17](https://modelscale.dev/model-profile?model=kimi-k2-5-reasoning)
- [Qwen3.8-Flash-Next (Alibaba) — 57.02](https://modelscale.dev/model-profile?model=qwen3-8-flash-next)
- [DeepSeek V3.2 (DeepSeek) — 56.87](https://modelscale.dev/model-profile?model=deepseek-v3-2)
- [GPT-5.2 Instant (OpenAI) — 56.83](https://modelscale.dev/model-profile?model=gpt-5-2-instant)
- [GPT-5.3 Instant (OpenAI) — 56.73](https://modelscale.dev/model-profile?model=gpt-5-3-instant)
- [GPT-5 (high) (OpenAI) — 56.47](https://modelscale.dev/model-profile?model=gpt-5-high)
- [Gemini 3.1 Flash-Lite (Google) — 56.32](https://modelscale.dev/model-profile?model=gemini-3-1-flash-lite)
- [Qwen3.5 Flash (Alibaba) — 56.2](https://modelscale.dev/model-profile?model=qwen3-5-flash)
- [DeepSeek V3.2 (Thinking) (DeepSeek) — 55.96](https://modelscale.dev/model-profile?model=deepseek-v3-2-thinking)
- [Qwen3 235B 2507 (Reasoning) (Alibaba) — 55.85](https://modelscale.dev/model-profile?model=qwen3-235b-2507-reasoning)
- [Hy3 Preview (Tencent) — 55.49](https://modelscale.dev/model-profile?model=hy3-preview)
- [GLM-4.5 (Z.AI) — 55.42](https://modelscale.dev/model-profile?model=glm-4-5)
- [Qwen3.5-27B (Alibaba) — 55.38](https://modelscale.dev/model-profile?model=qwen3-5-27b)
- [MiniMax M2.7 (MiniMax) — 55.14](https://modelscale.dev/model-profile?model=minimax-m2-7)
- [Gemma 4 26B A4B (Google) — 55.1](https://modelscale.dev/model-profile?model=gemma-4-26b-a4b)
- [Grok 4.1 Fast (Reasoning) (xAI) — 55.05](https://modelscale.dev/model-profile?model=grok-4-1-fast-reasoning)
- [MiniMax M2.5 (MiniMax) — 54.89](https://modelscale.dev/model-profile?model=minimax-m2-5)
- [Qwen3.5 397B (Alibaba) — 54.87](https://modelscale.dev/model-profile?model=qwen3-5-397b)
- [GPT-5.3-Codex-Spark (OpenAI) — 54.77](https://modelscale.dev/model-profile?model=gpt-5-3-codex-spark)
- [GPT-5.2-Codex (OpenAI) — 54.25](https://modelscale.dev/model-profile?model=gpt-5-2-codex)
- [Kimi K2.5 (Moonshot AI) — 54.14](https://modelscale.dev/model-profile?model=kimi-k2-5)
- [Claude Sonnet 4.5 (Anthropic) — 53.96](https://modelscale.dev/model-profile?model=claude-sonnet-4-5)
- [Qwen3 235B 2507 (Alibaba) — 53.89](https://modelscale.dev/model-profile?model=qwen3-235b-2507)
- [Trinity-Large-Preview (Arcee AI) — 53.81](https://modelscale.dev/model-profile?model=trinity-large-preview)
- [Qwen3.5-35B-A3B (Alibaba) — 53.57](https://modelscale.dev/model-profile?model=qwen3-5-35b-a3b)
- [Grok 4 (xAI) — 53.53](https://modelscale.dev/model-profile?model=grok-4)
- [Claude Opus 4.5 Thinking (Anthropic) — 53.04](https://modelscale.dev/model-profile?model=claude-opus-4-5-thinking)
- [Claude Haiku 4.5 (Anthropic) — 52.91](https://modelscale.dev/model-profile?model=claude-haiku-4-5)
- [Gemma 4 31B (Google) — 52.58](https://modelscale.dev/model-profile?model=gemma-4-31b)
- [DeepSeek LLM 2.0 (DeepSeek) — 52.43](https://modelscale.dev/model-profile?model=deepseek-llm-2-0)
- [GPT-5.1-Codex-Max (OpenAI) — 52.38](https://modelscale.dev/model-profile?model=gpt-5-1-codex-max)
- [MAI-Thinking-1 (Microsoft) — 52.01](https://modelscale.dev/model-profile?model=mai-thinking-1)
- [Grok 3 Beta (xAI) — 51.86](https://modelscale.dev/model-profile?model=grok-3-beta)
- [GLM-4.6 (Z.AI) — 51.86](https://modelscale.dev/model-profile?model=glm-4-6)
- [Apodex 1.1 (Apodex) — 51.74](https://modelscale.dev/model-profile?model=apodex-1-1)
- [Grok 4 Fast (Reasoning) (xAI) — 51.36](https://modelscale.dev/model-profile?model=grok-4-fast-reasoning)
- [Gemini 2.5 Flash (Google) — 51.3](https://modelscale.dev/model-profile?model=gemini-2-5-flash)
- [Nemotron 3 Nano 30B (NVIDIA) — 50.88](https://modelscale.dev/model-profile?model=nemotron-3-nano-30b)
- [Qwen3.5-122B-A10B (Alibaba) — 50.86](https://modelscale.dev/model-profile?model=qwen3-5-122b-a10b)
- [DeepSeek V3.1 (DeepSeek) — 50.84](https://modelscale.dev/model-profile?model=deepseek-v3-1)
- [DeepSeek-R1 (DeepSeek) — 50.62](https://modelscale.dev/model-profile?model=deepseek-r1)
- [Qwen3.5 Plus (Alibaba) — 50.11](https://modelscale.dev/model-profile?model=qwen3-5-plus)
- [Step 3.7 Flash (StepFun) — 49.98](https://modelscale.dev/model-profile?model=step-3-7-flash)
- [Step 3.5 Flash (StepFun) — 49.88](https://modelscale.dev/model-profile?model=step-3-5-flash)
- [GPT-5 (medium) (OpenAI) — 49.88](https://modelscale.dev/model-profile?model=gpt-5-medium)
- [Qwen3.7 Flash (Alibaba) — 49.24](https://modelscale.dev/model-profile?model=qwen3-7-flash)
- [MiMo-V2-Flash (Xiaomi) — 49.02](https://modelscale.dev/model-profile?model=mimo-v2-flash)
- [Nemotron 3 Super 120B A12B (NVIDIA) — 48.94](https://modelscale.dev/model-profile?model=nemotron-3-super-120b-a12b)
- [DeepSeek V3.1 (Reasoning) (DeepSeek) — 48.84](https://modelscale.dev/model-profile?model=deepseek-v3-1-reasoning)
- [DeepSeek Coder 2.0 (DeepSeek) — 48.26](https://modelscale.dev/model-profile?model=deepseek-coder-2-0)
- [Seed 1.6 (ByteDance) — 48.21](https://modelscale.dev/model-profile?model=seed-1-6)
- [GPT-5.1-Codex (OpenAI) — 48.1](https://modelscale.dev/model-profile?model=gpt-5-1-codex)
- [Nemotron 3 Super 100B (NVIDIA) — 48.07](https://modelscale.dev/model-profile?model=nemotron-3-super-100b)
- [o4-mini (high) (OpenAI) — 47.99](https://modelscale.dev/model-profile?model=o4-mini-high)
- [o3-pro (OpenAI) — 47.98](https://modelscale.dev/model-profile?model=o3-pro)
- [Qwen3.6-27B (Alibaba) — 47.89](https://modelscale.dev/model-profile?model=qwen3-6-27b)
- [DeepSeekMath V2 (DeepSeek) — 47.88](https://modelscale.dev/model-profile?model=deepseekmath-v2)
- [Qwen2.5-1M (Alibaba) — 47.88](https://modelscale.dev/model-profile?model=qwen2-5-1m)
- [Seed-2.0-Lite (ByteDance) — 47.83](https://modelscale.dev/model-profile?model=seed-2-0-lite)
- [o3 (OpenAI) — 47.71](https://modelscale.dev/model-profile?model=o3)
- [Ministral 3 14B (Reasoning) (Mistral) — 47.35](https://modelscale.dev/model-profile?model=ministral-3-14b-reasoning)
- [Grok 4.1 Fast (xAI) — 47.3](https://modelscale.dev/model-profile?model=grok-4-1-fast)
- [o3-mini (OpenAI) — 46.81](https://modelscale.dev/model-profile?model=o3-mini)
- [Grok 3 Mini (xAI) — 46.76](https://modelscale.dev/model-profile?model=grok-3-mini)
- [Ling 3.0 Flash (InclusionAI) — 46.75](https://modelscale.dev/model-profile?model=ling-3-0-flash)
- [Aion-2.0 (Aion Labs) — 46.67](https://modelscale.dev/model-profile?model=aion-2-0)
- [GLM-4.7-Flash (Z.AI) — 46.32](https://modelscale.dev/model-profile?model=glm-4-7-flash)
- [o1 (OpenAI) — 46.3](https://modelscale.dev/model-profile?model=o1)
- [o1-preview (OpenAI) — 46.2](https://modelscale.dev/model-profile?model=o1-preview)
- [GPT-OSS 120B (OpenAI) — 45.93](https://modelscale.dev/model-profile?model=gpt-oss-120b)
- [Mistral Large 3 (Mistral) — 45.6](https://modelscale.dev/model-profile?model=mistral-large-3)
- [GPT-5 mini (OpenAI) — 45.12](https://modelscale.dev/model-profile?model=gpt-5-mini)
- [Muse Glimmer 30B (Meta) — 45](https://modelscale.dev/model-profile?model=muse-glimmer-30b)
- [Qwen3.6-35B-A3B (Alibaba) — 43.99](https://modelscale.dev/model-profile?model=qwen3-6-35b-a3b)
- [GLM-4.5-Air (Z.AI) — 43.88](https://modelscale.dev/model-profile?model=glm-4-5-air)
- [Trinity-Large-Thinking (Arcee AI) — 43.43](https://modelscale.dev/model-profile?model=trinity-large-thinking)
- [Z-1 (Z) — 43.22](https://modelscale.dev/model-profile?model=z-1)
- [Seed 1.6 Flash (ByteDance) — 43.18](https://modelscale.dev/model-profile?model=seed-1-6-flash)
- [Mistral 8x7B (Mistral) — 43.08](https://modelscale.dev/model-profile?model=mistral-8x7b)
- [Mistral Small 4 (Mistral) — 43.01](https://modelscale.dev/model-profile?model=mistral-small-4)
- [Gemma 4 12B (Google) — 42.9](https://modelscale.dev/model-profile?model=gemma-4-12b)
- [Moonshot v1 (Moonshot AI) — 42.89](https://modelscale.dev/model-profile?model=moonshot-v1)
- [Command A+ (Cohere) — 42.75](https://modelscale.dev/model-profile?model=command-a-plus)
- [Seed-2.0-Mini (ByteDance) — 42.7](https://modelscale.dev/model-profile?model=seed-2-0-mini)
- [Nemotron Ultra 253B (NVIDIA) — 42.51](https://modelscale.dev/model-profile?model=nemotron-ultra-253b)
- [Mistral Medium 3.5 128B (Mistral) — 42.49](https://modelscale.dev/model-profile?model=mistral-medium-3-5-128b)
- [Mercury 2 (Inception) — 42.33](https://modelscale.dev/model-profile?model=mercury-2)
- [Claude 4.1 Opus (Anthropic) — 42.02](https://modelscale.dev/model-profile?model=claude-4-1-opus)
- [o1-pro (OpenAI) — 41.94](https://modelscale.dev/model-profile?model=o1-pro)
- [DeepSeek V3 (DeepSeek) — 41.74](https://modelscale.dev/model-profile?model=deepseek-v3)
- [GPT-4.1 (OpenAI) — 41.39](https://modelscale.dev/model-profile?model=gpt-4-1)
- [GPT-OSS 20B (OpenAI) — 41.18](https://modelscale.dev/model-profile?model=gpt-oss-20b)
- [Qwen3 Max (Alibaba) — 40.87](https://modelscale.dev/model-profile?model=qwen3-max)
- [Nemotron 3 Nano Omni 30B A3B (NVIDIA) — 40.77](https://modelscale.dev/model-profile?model=nemotron-3-nano-omni-30b-a3b)
- [Ling 2.6 Flash (InclusionAI) — 40.47](https://modelscale.dev/model-profile?model=ling-2-6-flash)
- [Nemotron 3 Ultra (NVIDIA) — 40.39](https://modelscale.dev/model-profile?model=nemotron-3-ultra-500b)
- [Gemma 4 E4B (Google) — 40.06](https://modelscale.dev/model-profile?model=gemma-4-e4b)
- [Sarvam 105B (Sarvam) — 39.99](https://modelscale.dev/model-profile?model=sarvam-105b)
- [LFM2.5-2.6B (LiquidAI) — 39.77](https://modelscale.dev/model-profile?model=lfm2-5-2-6b)
- [Gemma 4 E2B (Google) — 39.45](https://modelscale.dev/model-profile?model=gemma-4-e2b)
- [Claude 4 Sonnet (Anthropic) — 39.4](https://modelscale.dev/model-profile?model=claude-4-sonnet)
- [LFM2.5-8B-A1B (LiquidAI) — 39.15](https://modelscale.dev/model-profile?model=lfm2-5-8b-a1b)
- [GPT-4o (OpenAI) — 38.95](https://modelscale.dev/model-profile?model=gpt-4o)
- [Sarvam 30B (Sarvam) — 38.8](https://modelscale.dev/model-profile?model=sarvam-30b)
- [Ministral 3 8B (Reasoning) (Mistral) — 38.58](https://modelscale.dev/model-profile?model=ministral-3-8b-reasoning)
- [Qwen3-Omni-30B-A3B-Instruct (Alibaba) — 38.5](https://modelscale.dev/model-profile?model=qwen3-omni-30b-a3b-instruct)
- [Granite 4.2 8B (IBM) — 38.44](https://modelscale.dev/model-profile?model=granite-4-2-8b)
- [Granite-4.0-H-1B (IBM) — 38.07](https://modelscale.dev/model-profile?model=granite-4-0-h-1b)
- [Qwen2.5-VL-32B (Alibaba) — 38.06](https://modelscale.dev/model-profile?model=qwen2-5-vl-32b)
- [Llama 4 Behemoth (Meta) — 38.01](https://modelscale.dev/model-profile?model=llama-4-behemoth)
- [Granite-4.0-1B (IBM) — 37.97](https://modelscale.dev/model-profile?model=granite-4-0-1b)
- [Granite-4.0-350M (IBM) — 37.87](https://modelscale.dev/model-profile?model=granite-4-0-350m)
- [Granite-4.0-H-350M (IBM) — 37.87](https://modelscale.dev/model-profile?model=granite-4-0-h-350m)
- [Ministral 3 3B (Reasoning) (Mistral) — 37.73](https://modelscale.dev/model-profile?model=ministral-3-3b-reasoning)
- [Qwen2.5-72B (Alibaba) — 37.55](https://modelscale.dev/model-profile?model=qwen2-5-72b)
- [Mistral 8x7B v0.2 (Mistral) — 37.36](https://modelscale.dev/model-profile?model=mistral-8x7b-v0-2)
- [GPT-5 nano (OpenAI) — 37.01](https://modelscale.dev/model-profile?model=gpt-5-nano)
- [Ornith-1.5-35B-A3B (Ornith AI) — 36.79](https://modelscale.dev/model-profile?model=ornith-1-5-35b-a3b)
- [Llama 3.1 405B (Meta) — 36.73](https://modelscale.dev/model-profile?model=llama-3-1-405b)
- [Mistral Large 2 (Mistral) — 36.47](https://modelscale.dev/model-profile?model=mistral-large-2)
- [Gemma 3 27B (Google) — 35.82](https://modelscale.dev/model-profile?model=gemma-3-27b)
- [Claude 3 Opus (Anthropic) — 34.52](https://modelscale.dev/model-profile?model=claude-3-opus)
- [Grok Code Fast 1 (xAI) — 34.21](https://modelscale.dev/model-profile?model=grok-code-fast-1)
- [Gemini 1.5 Pro (Google) — 33.6](https://modelscale.dev/model-profile?model=gemini-1-5-pro)
- [GPT-4.1 mini (OpenAI) — 33.47](https://modelscale.dev/model-profile?model=gpt-4-1-mini)
- [Qwen2.5 Coder 32B Instruct (Alibaba) — 33.43](https://modelscale.dev/model-profile?model=qwen2-5-coder-32b-instruct)
- [Llama 4 Scout (Meta) — 33.42](https://modelscale.dev/model-profile?model=llama-4-scout)
- [GPT-4o mini (OpenAI) — 33.37](https://modelscale.dev/model-profile?model=gpt-4o-mini)
- [Phi-4 (Microsoft) — 33.18](https://modelscale.dev/model-profile?model=phi-4)
- [DeepSeek R1 Distill Qwen 32B (DeepSeek) — 32.62](https://modelscale.dev/model-profile?model=deepseek-r1-distill-qwen-32b)
- [Claude 3.5 Sonnet (Anthropic) — 32.12](https://modelscale.dev/model-profile?model=claude-3-5-sonnet)
- [Claude 4.1 Opus Thinking (Anthropic) — 31.67](https://modelscale.dev/model-profile?model=claude-4-1-opus-thinking)
- [Llama 3 70B (Meta) — 30.76](https://modelscale.dev/model-profile?model=llama-3-70b)
- [Mistral Medium 3 (Mistral) — 30.72](https://modelscale.dev/model-profile?model=mistral-medium-3)
- [Ministral 3 14B (Mistral) — 30.32](https://modelscale.dev/model-profile?model=ministral-3-14b)
- [GPT-4.1 nano (OpenAI) — 27.31](https://modelscale.dev/model-profile?model=gpt-4-1-nano)
- [Mixtral 8x22B Instruct v0.1 (Mistral) — 26.88](https://modelscale.dev/model-profile?model=mixtral-8x22b-instruct-v0-1)
- [GPT-4 Turbo (OpenAI) — 24.52](https://modelscale.dev/model-profile?model=gpt-4-turbo)
- [Kimi K2 (Moonshot AI) — 23.31](https://modelscale.dev/model-profile?model=kimi-k2)
- [Llama 4 Maverick (Meta) — 22.86](https://modelscale.dev/model-profile?model=llama-4-maverick)
- [MiniMax M1 80k (MiniMax) — 21.56](https://modelscale.dev/model-profile?model=minimax-m1-80k)
- [Nemotron 3.5 Lightning 30B A3B NVFP4 (NVIDIA) — 21.02](https://modelscale.dev/model-profile?model=nemotron-3-5-lightning-30b-a3b-nvfp4)
- [Nova Pro (Amazon) — 19.18](https://modelscale.dev/model-profile?model=nova-pro)
- [LFM2-24B-A2B (LiquidAI) — 18.61](https://modelscale.dev/model-profile?model=lfm2-24b-a2b)
- [Claude 3 Haiku (Anthropic) — 18.1](https://modelscale.dev/model-profile?model=claude-3-haiku)
- [Ministral 3 8B (Mistral) — 17.92](https://modelscale.dev/model-profile?model=ministral-3-8b)
- [LFM2.5-1.2B-Thinking (LiquidAI) — 16.78](https://modelscale.dev/model-profile?model=lfm2-5-1-2b-thinking)
- [Gemini 1.0 Pro (Google) — 16.46](https://modelscale.dev/model-profile?model=gemini-1-0-pro)
- [LFM2.5-1.2B-Instruct (LiquidAI) — 16.1](https://modelscale.dev/model-profile?model=lfm2-5-1-2b-instruct)
- [Ministral 3 3B (Mistral) — 16.09](https://modelscale.dev/model-profile?model=ministral-3-3b)
- [Nemotron-4 15B (NVIDIA) — 15.92](https://modelscale.dev/model-profile?model=nemotron-4-15b)
- [Mistral 7B v0.3 (Mistral) — 8.68](https://modelscale.dev/model-profile?model=mistral-7b-v0-3)
- [Laguna M.1 (Poolside) — 3.88](https://modelscale.dev/model-profile?model=laguna-m-1)
- [Laguna XS.2 (Poolside) — 1.51](https://modelscale.dev/model-profile?model=laguna-xs-2)
- [1-bit Bonsai 1.7B (Prism ML)](https://modelscale.dev/model-profile?model=bonsai-1-7b)
- [1-bit Bonsai 27B (Prism ML)](https://modelscale.dev/model-profile?model=bonsai-27b)
- [1-bit Bonsai 4B (Prism ML)](https://modelscale.dev/model-profile?model=bonsai-4b)
- [1-bit Bonsai 8B (Prism ML)](https://modelscale.dev/model-profile?model=bonsai-8b)
- [A.X K2 (SK Telecom)](https://modelscale.dev/model-profile?model=a-x-k2)
- [A.X series (SK Telecom)](https://modelscale.dev/model-profile?model=skt-ax)
- [Agents-A1 (InternScience)](https://modelscale.dev/model-profile?model=agents-a1)
- [Agents-A1-4B (InternScience)](https://modelscale.dev/model-profile?model=agents-a1-4b)
- [Agents-A1-F16-GGUF (InternScience)](https://modelscale.dev/model-profile?model=agents-a1-f16-gguf)
- [Agents-A1-FP8 (InternScience)](https://modelscale.dev/model-profile?model=agents-a1-fp8)
- [Agents-A1-Q4_K_M-GGUF (InternScience)](https://modelscale.dev/model-profile?model=agents-a1-q4-k-m-gguf)
- [Agents-A1-Q8_0-GGUF (InternScience)](https://modelscale.dev/model-profile?model=agents-a1-q8-0-gguf)
- [Amazon Nova 2 Lite (Amazon)](https://modelscale.dev/model-profile?model=nova-2-lite)
- [Amazon Nova 2 Pro (Amazon)](https://modelscale.dev/model-profile?model=nova-2-pro)
- [Amazon Nova 2 Sonic (Amazon)](https://modelscale.dev/model-profile?model=amazon-nova-2-sonic)
- [Apodex 1.1 Mini (Apodex)](https://modelscale.dev/model-profile?model=apodex-1-1-mini)
- [Atria Dawn Preview (Shanghai Artificial Intelligence Laboratory)](https://modelscale.dev/model-profile?model=atria-dawn-preview)
- [Audio Flamingo 3 7B (NVIDIA)](https://modelscale.dev/model-profile?model=audio-flamingo-3-7b)
- [AuK (Tencent)](https://modelscale.dev/model-profile?model=tencent-auk)
- [Baichuan-Audio (Baichuan)](https://modelscale.dev/model-profile?model=baichuan-audio)
- [Baichuan-Omni 1.5 (Baichuan)](https://modelscale.dev/model-profile?model=baichuan-omni-1-5)
- [Bland Speech v3 (Bland AI)](https://modelscale.dev/model-profile?model=bland-speech-v3)
- [BLSP 7B (BLSP authors)](https://modelscale.dev/model-profile?model=blsp-7b)
- [BR-Voice-Reasoner (Bairong Xdynamics)](https://modelscale.dev/model-profile?model=br-voice-reasoner)
- [BTL-3 (Bad Theory Labs)](https://modelscale.dev/model-profile?model=btl-3)
- [BTL-4 (Bad Theory Labs)](https://modelscale.dev/model-profile?model=btl-4)
- [BTL-4 Compact (Bad Theory Labs)](https://modelscale.dev/model-profile?model=btl-4-compact)
- [Cartesia Ink-Whisper (Cartesia)](https://modelscale.dev/model-profile?model=cartesia-ink-whisper)
- [Cartesia Sonic 3 (Cartesia)](https://modelscale.dev/model-profile?model=cartesia-sonic-3)
- [Cartesia Sonic 3.5 (Cartesia)](https://modelscale.dev/model-profile?model=cartesia-sonic-3-5)
- [Cartesia Sonic 3.6 (Cartesia)](https://modelscale.dev/model-profile?model=cartesia-sonic-3-6)
- [Celeris-1 (Celeris)](https://modelscale.dev/model-profile?model=celeris-1)
- [Claude Haiku 4.5 Thinking (Anthropic)](https://modelscale.dev/model-profile?model=claude-haiku-4-5-thinking)
- [Claude Mythos 5 (Anthropic)](https://modelscale.dev/model-profile?model=claude-mythos-5)
- [Claude Mythos 5.1 (Anthropic)](https://modelscale.dev/model-profile?model=claude-mythos-5-1)
- [Claude Mythos Preview (Anthropic)](https://modelscale.dev/model-profile?model=claude-mythos-preview)
- [Claude Sonnet 4.5 Thinking (Anthropic)](https://modelscale.dev/model-profile?model=claude-sonnet-4-5-thinking)
- [Cohere Transcribe 03-2026 (Cohere)](https://modelscale.dev/model-profile?model=cohere-transcribe-03-2026)
- [Cohere Transcribe Arabic 07-2026 (Cohere)](https://modelscale.dev/model-profile?model=cohere-transcribe-arabic-07-2026)
- [Composer 2 (Cursor)](https://modelscale.dev/model-profile?model=composer-2)
- [Composer 2 Fast (Cursor)](https://modelscale.dev/model-profile?model=composer-2-fast)
- [Composer 2.5 (Cursor)](https://modelscale.dev/model-profile?model=composer-2-5)
- [Cosmos3-Edge (NVIDIA)](https://modelscale.dev/model-profile?model=cosmos3-edge)
- [DBRX Instruct (Databricks)](https://modelscale.dev/model-profile?model=dbrx-instruct)
- [Deepgram Aura-2 (Deepgram)](https://modelscale.dev/model-profile?model=deepgram-aura-2)
- [Deepgram Flux (Deepgram)](https://modelscale.dev/model-profile?model=deepgram-flux)
- [Deepgram Flux Multilingual (Deepgram)](https://modelscale.dev/model-profile?model=deepgram-flux-multilingual)
- [Deepgram Nova-3 (Deepgram)](https://modelscale.dev/model-profile?model=deepgram-nova-3)
- [DeepSeek V4 Flash 0731 (DeepSeek)](https://modelscale.dev/model-profile?model=deepseek-v4-flash-max)
- [DeepSeek V4.1 Flash (DeepSeek)](https://modelscale.dev/model-profile?model=deepseek-v4-1-flash)
- [DiVA Llama 3 8B (DiVA authors)](https://modelscale.dev/model-profile?model=diva-llama-3-v0-8b)
- [DNA 1.0 8B (Community)](https://modelscale.dev/model-profile?model=dna-1-0)
- [ElevenLabs Eleven v3 (ElevenLabs)](https://modelscale.dev/model-profile?model=elevenlabs-eleven-v3)
- [ElevenLabs Eleven v3 Conversational (ElevenLabs)](https://modelscale.dev/model-profile?model=elevenlabs-eleven-v3-conversational)
- [Exaone 4.0 1.2B (LG AI Research)](https://modelscale.dev/model-profile?model=exaone-4-0-1-2b)
- [Exaone 4.0 32B (LG AI Research)](https://modelscale.dev/model-profile?model=exaone-4-0-32b)
- [Fara1.5-27B (Microsoft)](https://modelscale.dev/model-profile?model=fara-1-5-27b)
- [Fara1.5-4B (Microsoft)](https://modelscale.dev/model-profile?model=fara-1-5-4b)
- [Fugu Cyber (Sakana AI)](https://modelscale.dev/model-profile?model=sakana-fugu-cyber)
- [Gemini 2.5 Flash Native Audio Preview (12-2025) (Google)](https://modelscale.dev/model-profile?model=gemini-2-5-flash-native-audio-preview-12-2025)
- [Gemini 2.5 Flash TTS Preview (Google)](https://modelscale.dev/model-profile?model=gemini-2-5-flash-tts-preview)
- [Gemini 2.5 Pro TTS Preview (Google)](https://modelscale.dev/model-profile?model=gemini-2-5-pro-tts-preview)
- [Gemini 3.1 Flash Live Preview (Google)](https://modelscale.dev/model-profile?model=gemini-3-1-flash-live-preview)
- [Gemini 3.1 Flash TTS Preview (Google)](https://modelscale.dev/model-profile?model=gemini-3-1-flash-tts-preview)
- [Gemini 3.5 Flash Cyber (Google)](https://modelscale.dev/model-profile?model=gemini-3-5-flash-cyber)
- [Gemini 3.5 Transcribe (Google)](https://modelscale.dev/model-profile?model=gemini-3-5-transcribe)
- [Gemini 3.8 Flash Cyber (Google)](https://modelscale.dev/model-profile?model=gemini-3-8-flash-cyber)
- [Gemini 3.8 Live (Google)](https://modelscale.dev/model-profile?model=gemini-3-8-live)
- [Gemini 3.8 Live Extended Thinking (Google)](https://modelscale.dev/model-profile?model=gemini-3-8-live-extended-thinking)
- [GLM Realtime Air (Zhipu AI)](https://modelscale.dev/model-profile?model=glm-realtime-air)
- [GLM Realtime Flash (Zhipu AI)](https://modelscale.dev/model-profile?model=glm-realtime-flash)
- [GLM-4-Voice (Zhipu AI)](https://modelscale.dev/model-profile?model=glm-4-voice)
- [GPT Audio 1.5 (OpenAI)](https://modelscale.dev/model-profile?model=gpt-audio-1-5)
- [GPT Live Transcribe (OpenAI)](https://modelscale.dev/model-profile?model=gpt-live-transcribe)
- [GPT Realtime (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime)
- [GPT Realtime 1.5 (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime-1-5)
- [GPT Realtime 2 (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime-2)
- [GPT Realtime 2.1 (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime-2-1)
- [GPT Realtime 2.1 Mini (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime-2-1-mini)
- [GPT Realtime mini (OpenAI)](https://modelscale.dev/model-profile?model=gpt-realtime-mini)
- [GPT Transcribe (OpenAI)](https://modelscale.dev/model-profile?model=gpt-transcribe)
- [GPT-4o Audio (OpenAI)](https://modelscale.dev/model-profile?model=gpt-4o-audio)
- [GPT-4o mini Audio (OpenAI)](https://modelscale.dev/model-profile?model=gpt-4o-mini-audio)
- [GPT-4o mini TTS (OpenAI)](https://modelscale.dev/model-profile?model=gpt-4o-mini-tts)
- [GPT-5.6 Cyber (OpenAI)](https://modelscale.dev/model-profile?model=gpt-5-6-cyber)
- [GPT-Live-1 (OpenAI)](https://modelscale.dev/model-profile?model=gpt-live-1)
- [Granite 4.2 30B (IBM)](https://modelscale.dev/model-profile?model=granite-4-2-30b)
- [Granite 4.2 3B (IBM)](https://modelscale.dev/model-profile?model=granite-4-2-3b)
- [Granite Speech 5.0 470M TurboCTC (IBM)](https://modelscale.dev/model-profile?model=granite-speech-5-0-470m-turboctc)
- [Grok 4.20 Multi-agent (xAI)](https://modelscale.dev/model-profile?model=grok-4-20-multi-agent-beta)
- [Grok Build 0.1 (xAI)](https://modelscale.dev/model-profile?model=grok-build-0-1)
- [Grok Realtime (xAI)](https://modelscale.dev/model-profile?model=grok-realtime)
- [Grok TTS (xAI)](https://modelscale.dev/model-profile?model=grok-tts)
- [Grok Voice Think Fast 1.0 (xAI)](https://modelscale.dev/model-profile?model=grok-voice-think-fast-1-0)
- [Grok Voice Think Fast 2.0 (xAI)](https://modelscale.dev/model-profile?model=grok-voice-think-fast-2-0)
- [Hark Handoff (Hark)](https://modelscale.dev/model-profile?model=hark-handoff)
- [Holo2-235B-A22B (H Company)](https://modelscale.dev/model-profile?model=holo2-235b-a22b)
- [Holo2-30B-A3B (H Company)](https://modelscale.dev/model-profile?model=holo2-30b-a3b)
- [Holo2-4B (H Company)](https://modelscale.dev/model-profile?model=holo2-4b)
- [Holo2-8B (H Company)](https://modelscale.dev/model-profile?model=holo2-8b)
- [Holo3-122B-A10B (H Company)](https://modelscale.dev/model-profile?model=holo3-122b-a10b)
- [Holo3-35B-A3B (H Company)](https://modelscale.dev/model-profile?model=holo3-35b-a3b)
- [Holo3.1-0.8B (H Company)](https://modelscale.dev/model-profile?model=holo3-1-0-8b)
- [Holo3.1-35B-A3B (H Company)](https://modelscale.dev/model-profile?model=holo3-1-35b-a3b)
- [Holo3.1-35B-A3B-FP8 (H Company)](https://modelscale.dev/model-profile?model=holo3-1-35b-a3b-fp8)
- [Holo3.1-35B-A3B-GGUF (H Company)](https://modelscale.dev/model-profile?model=holo3-1-35b-a3b-gguf)
- [Holo3.1-35B-A3B-NVFP4 (H Company)](https://modelscale.dev/model-profile?model=holo3-1-35b-a3b-nvfp4)
- [Holo3.1-4B (H Company)](https://modelscale.dev/model-profile?model=holo3-1-4b)
- [Holo3.1-9B (H Company)](https://modelscale.dev/model-profile?model=holo3-1-9b)
- [Holotron-12B (H Company)](https://modelscale.dev/model-profile?model=holotron-12b)
- [Hy ASR 3.0 preview (Tencent)](https://modelscale.dev/model-profile?model=hy-asr-3-0-preview)
- [Hy-MT1.5-1.8B-1.25bit (Tencent Hunyuan)](https://modelscale.dev/model-profile?model=hy-mt1-5-1-8b-1-25bit)
- [Hy-MT2-1.8B (Tencent)](https://modelscale.dev/model-profile?model=hy-mt2-1-8b)
- [Hy-MT2-30B-A3B (Tencent)](https://modelscale.dev/model-profile?model=hy-mt2-30b-a3b)
- [HyperClova X Dash (Naver Cloud)](https://modelscale.dev/model-profile?model=hyperclova-x-dash)
- [HyperClova X Think 32B (Naver Cloud)](https://modelscale.dev/model-profile?model=hyperclova-x-think)
- [Ichigo (Jan)](https://modelscale.dev/model-profile?model=ichigo)
- [Interfaze Beta (Interfaze)](https://modelscale.dev/model-profile?model=interfaze-beta)
- [Inworld Realtime TTS-2 (Inworld)](https://modelscale.dev/model-profile?model=inworld-tts-2)
- [Inworld Realtime TTS-2 Flash (Inworld)](https://modelscale.dev/model-profile?model=inworld-tts-2-flash)
- [Inworld TTS-1.5 Max (Inworld)](https://modelscale.dev/model-profile?model=inworld-tts-1-5-max)
- [K-Exaone (LG AI Research)](https://modelscale.dev/model-profile?model=k-exaone)
- [K-EXAONE 2.0 (LG AI Research)](https://modelscale.dev/model-profile?model=k-exaone-2-0)
- [Kalpa TTS Beta v0.1 (Kalpa Labs)](https://modelscale.dev/model-profile?model=kalpa-tts-beta-v0-1)
- [Kanana Essence (Kakao)](https://modelscale.dev/model-profile?model=kanana-essence)
- [Kanana Flag (Kakao)](https://modelscale.dev/model-profile?model=kanana-flag)
- [Kanana Nano (Kakao)](https://modelscale.dev/model-profile?model=kanana-nano)
- [Kanana-2 1.3B Instruct (Kakao)](https://modelscale.dev/model-profile?model=kanana-2-1-3b-instruct)
- [Kanana-2 3B Instruct (Kakao)](https://modelscale.dev/model-profile?model=kanana-2-3b-instruct)
- [Kimi-Audio 7B (Moonshot AI)](https://modelscale.dev/model-profile?model=kimi-audio-7b)
- [Kokoro 82M (hexgrad)](https://modelscale.dev/model-profile?model=kokoro-82m)
- [Kyutai Pocket TTS (Kyutai)](https://modelscale.dev/model-profile?model=kyutai-pocket-tts)
- [Laguna S 2.1 (Poolside)](https://modelscale.dev/model-profile?model=laguna-s-2-1)
- [Laguna XS 2.1 (Poolside)](https://modelscale.dev/model-profile?model=laguna-xs-2-1)
- [Leanstral (Mistral)](https://modelscale.dev/model-profile?model=leanstral)
- [Leanstral 1.5 (Mistral)](https://modelscale.dev/model-profile?model=leanstral-1-5)
- [LFG-1 (glenn2)](https://modelscale.dev/model-profile?model=lfg-1)
- [LFG-2 (glenn2)](https://modelscale.dev/model-profile?model=lfg-2)
- [LFG-3 (glenn2)](https://modelscale.dev/model-profile?model=lfg-3)
- [LFM2.5-1.2B-JP-202606 (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-1-2b-jp-202606)
- [LFM2.5-230M (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-230m)
- [LFM2.5-350M (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-350m)
- [LFM2.5-ColBERT-350M (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-colbert-350m)
- [LFM2.5-Embedding-350M (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-embedding-350m)
- [LFM2.5-VL-1.6B-Extract (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-vl-1-6b-extract)
- [LFM2.5-VL-3B (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-vl-3b)
- [LFM2.5-VL-450M (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-vl-450m)
- [LFM2.5-VL-450M-Extract (LiquidAI)](https://modelscale.dev/model-profile?model=lfm2-5-vl-450m-extract)
- [Lightning v3.1 Pro (Smallest AI)](https://modelscale.dev/model-profile?model=lightning-v3-1-pro)
- [Ling 3.0 Flash Fin (InclusionAI)](https://modelscale.dev/model-profile?model=ling-3-0-flash-fin)
- [Ling 3.0 Flash FP8 (InclusionAI)](https://modelscale.dev/model-profile?model=ling-3-0-flash-fp8)
- [Ling 3.0 Flash VL (InclusionAI)](https://modelscale.dev/model-profile?model=ling-3-0-flash-vl)
- [Ling 3.0 Tiny (InclusionAI)](https://modelscale.dev/model-profile?model=ling-3-0-tiny)
- [LLaDA2.2-flash (InclusionAI)](https://modelscale.dev/model-profile?model=llada2-2-flash)
- [LLaDA2.2-mini (InclusionAI)](https://modelscale.dev/model-profile?model=llada2-2-mini)
- [LLaMA-Omni (ICTNLP)](https://modelscale.dev/model-profile?model=llama-omni)
- [LongCat-2.0 (Meituan)](https://modelscale.dev/model-profile?model=longcat-2-0)
- [LongCat-Flash-Lite-Sparse (Meituan)](https://modelscale.dev/model-profile?model=longcat-flash-lite-sparse)
- [Lyra Base (DVL Lab)](https://modelscale.dev/model-profile?model=lyra-base)
- [Lyra Mini (DVL Lab)](https://modelscale.dev/model-profile?model=lyra-mini)
- [Macaw (Bad Theory Labs)](https://modelscale.dev/model-profile?model=macaw-v1)
- [MAI-Code-1.1-Flash (Microsoft)](https://modelscale.dev/model-profile?model=mai-code-1-1-flash)
- [MAI-Cyber-1-Flash (Microsoft)](https://modelscale.dev/model-profile?model=mai-cyber-1-flash)
- [MAI-Transcribe-1.5 (Microsoft)](https://modelscale.dev/model-profile?model=mai-transcribe-1-5)
- [MAI-Transcribe-2 (Microsoft)](https://modelscale.dev/model-profile?model=mai-transcribe-2)
- [MAI-Voice-2 (Microsoft)](https://modelscale.dev/model-profile?model=mai-voice-2)
- [MAI-Voice-2-Flash (Microsoft)](https://modelscale.dev/model-profile?model=mai-voice-2-flash)
- [Mair-hub 0.5B Omni (Mair Hub)](https://modelscale.dev/model-profile?model=mair-hub-0-5b-omni)
- [Megrez-3B-Omni (Infinigence AI)](https://modelscale.dev/model-profile?model=megrez-3b-omni)
- [Mellum2-12B-A2.5B-Instruct (JetBrains)](https://modelscale.dev/model-profile?model=mellum2-12b-a2-5b-instruct)
- [Mellum2-12B-A2.5B-Thinking (JetBrains)](https://modelscale.dev/model-profile?model=mellum2-12b-a2-5b-thinking)
- [MERaLiON-AudioLLM (AI Singapore)](https://modelscale.dev/model-profile?model=meralion-audiollm)
- [Mercury 2.5 (Inception)](https://modelscale.dev/model-profile?model=mercury-2-5)
- [Mercury 2.5 Preview (Inception)](https://modelscale.dev/model-profile?model=mercury-2-5-preview)
- [Mercury Edit 2 (Inception)](https://modelscale.dev/model-profile?model=mercury-edit-2)
- [Mini-Omni 0.5B (gpt-omni)](https://modelscale.dev/model-profile?model=mini-omni)
- [Mini-Omni2 (gpt-omni)](https://modelscale.dev/model-profile?model=mini-omni2)
- [MiniCPM-o 2.6 (OpenBMB)](https://modelscale.dev/model-profile?model=minicpm-o-2-6)
- [MiniCPM5-1B (OpenBMB)](https://modelscale.dev/model-profile?model=minicpm5-1b)
- [MiniCPM5-2B (OpenBMB)](https://modelscale.dev/model-profile?model=minicpm5-2b)
- [MiniMax Speech-02 HD (MiniMax)](https://modelscale.dev/model-profile?model=minimax-speech-02-hd)
- [Mistral Small 4 (Reasoning) (Mistral)](https://modelscale.dev/model-profile?model=mistral-small-4-reasoning)
- [Moshi 7B (Kyutai)](https://modelscale.dev/model-profile?model=moshi-7b)
- [Muse Spark 1.3 (Meta)](https://modelscale.dev/model-profile?model=muse-spark-1-3)
- [Muse Voice Transcribe (Meta)](https://modelscale.dev/model-profile?model=muse-voice-transcribe)
- [NExT-GPT 7B (NExT-GPT authors)](https://modelscale.dev/model-profile?model=next-gpt-7b)
- [North Micro Vision Instruct (Cohere)](https://modelscale.dev/model-profile?model=north-micro-vision-instruct)
- [North Mini Code (Cohere)](https://modelscale.dev/model-profile?model=north-mini-code-1-0)
- [North Small Translate (Cohere)](https://modelscale.dev/model-profile?model=north-small-translate-1-0)
- [Ola (Ola-Omni)](https://modelscale.dev/model-profile?model=ola-omni)
- [OriOn-Mistral-24B (LightOn)](https://modelscale.dev/model-profile?model=orion-mistral-24b)
- [OriOn-Qwen-32B (LightOn)](https://modelscale.dev/model-profile?model=orion-qwen-32b)
- [Ornith-1.0-35B (DeepReinforce AI)](https://modelscale.dev/model-profile?model=ornith-1-0-35b)
- [Ornith-1.0-397B (DeepReinforce AI)](https://modelscale.dev/model-profile?model=ornith-1-0-397b)
- [Ornith-1.0-9B (DeepReinforce AI)](https://modelscale.dev/model-profile?model=ornith-1-0-9b)
- [Ornith-1.5-9B (Ornith AI)](https://modelscale.dev/model-profile?model=ornith-1-5-9b)
- [PandaGPT 7B (PandaGPT authors)](https://modelscale.dev/model-profile?model=pandagpt-7b)
- [Parakeet CTC 1.1B (NVIDIA)](https://modelscale.dev/model-profile?model=parakeet-ctc-1-1b)
- [Pharia-1-LLM-7B-control (Aleph Alpha)](https://modelscale.dev/model-profile?model=pharia-1-llm-7b-control)
- [Pharia-1-LLM-7B-control-aligned (Aleph Alpha)](https://modelscale.dev/model-profile?model=pharia-1-llm-7b-control-aligned)
- [Phi-4 Multimodal Instruct (Microsoft)](https://modelscale.dev/model-profile?model=phi-4-multimodal-instruct)
- [Pokee-Isaac 28B (Pokee AI)](https://modelscale.dev/model-profile?model=pokee-isaac-28b)
- [Quasar 438B (Multiverse Computing)](https://modelscale.dev/model-profile?model=quasar-438b)
- [Qwen-AgentWorld-35B-A3B (Alibaba)](https://modelscale.dev/model-profile?model=qwen-agentworld-35b-a3b)
- [Qwen-Audio 7B (Alibaba)](https://modelscale.dev/model-profile?model=qwen-audio-7b)
- [Qwen-Audio-Chat 7B (Alibaba)](https://modelscale.dev/model-profile?model=qwen-audio-chat-7b)
- [Qwen2-Audio 7B Instruct (Alibaba)](https://modelscale.dev/model-profile?model=qwen2-audio-7b-instruct)
- [Qwen2.5-Omni 7B (Alibaba)](https://modelscale.dev/model-profile?model=qwen2-5-omni-7b)
- [Qwen3-ASR 0.6B (Alibaba)](https://modelscale.dev/model-profile?model=qwen3-asr-0-6b)
- [Qwen3-ASR 1.7B (Alibaba)](https://modelscale.dev/model-profile?model=qwen3-asr-1-7b)
- [Qwen3-Omni-30B-A3B-Thinking (Alibaba)](https://modelscale.dev/model-profile?model=qwen3-omni-30b-a3b-thinking)
- [Qwen3.8 Max Preview (Alibaba)](https://modelscale.dev/model-profile?model=qwen3-8-max-preview)
- [Sakana Fugu (Sakana AI)](https://modelscale.dev/model-profile?model=sakana-fugu)
- [Sakana Fugu-Ultra (Sakana AI)](https://modelscale.dev/model-profile?model=sakana-fugu-ultra)
- [Sakana Fugu-Ultra v1.1 (Sakana AI)](https://modelscale.dev/model-profile?model=sakana-fugu-ultra-v1-1)
- [Sakana Namazu (Sakana AI)](https://modelscale.dev/model-profile?model=sakana-namazu)
- [SALMONN 13B (SALMONN authors)](https://modelscale.dev/model-profile?model=salmonn-13b)
- [SALMONN 7B (SALMONN authors)](https://modelscale.dev/model-profile?model=salmonn-7b)
- [Scribe v2.2 Realtime (ElevenLabs)](https://modelscale.dev/model-profile?model=scribe-v2-2-realtime)
- [Seed 2.1 Pro (ByteDance)](https://modelscale.dev/model-profile?model=seed-2-1-pro)
- [Seed 2.1 Turbo (ByteDance)](https://modelscale.dev/model-profile?model=seed-2-1-turbo)
- [SeedRealtime (ByteDance)](https://modelscale.dev/model-profile?model=seedrealtime)
- [SLAM-Omni (X-LANCE)](https://modelscale.dev/model-profile?model=slam-omni)
- [Solar Open 2 (Upstage)](https://modelscale.dev/model-profile?model=solar-open-2)
- [Solar Pro 2 (Upstage)](https://modelscale.dev/model-profile?model=solar-pro-2)
- [Solar Pro 3 (Upstage)](https://modelscale.dev/model-profile?model=solar-pro-3)
- [Solar Pro 4 (Upstage)](https://modelscale.dev/model-profile?model=solar-pro-4)
- [Soofi S 30B-A3B (Soofi Project)](https://modelscale.dev/model-profile?model=soofi-s-30b-a3b)
- [SpeechGPT 7B (OpenMOSS)](https://modelscale.dev/model-profile?model=speechgpt-7b)
- [Step-Audio (StepFun)](https://modelscale.dev/model-profile?model=step-audio)
- [Step-Audio-Chat 130B (StepFun)](https://modelscale.dev/model-profile?model=step-audio-chat-130b)
- [SWE-1.7 (Cognition)](https://modelscale.dev/model-profile?model=swe-1-7)
- [SWE-2 (Cognition)](https://modelscale.dev/model-profile?model=swe-2)
- [Ternary Bonsai 1.7B (Prism ML)](https://modelscale.dev/model-profile?model=ternary-bonsai-1-7b)
- [Ternary Bonsai 27B (Prism ML)](https://modelscale.dev/model-profile?model=ternary-bonsai-27b)
- [Ternary Bonsai 4B (Prism ML)](https://modelscale.dev/model-profile?model=ternary-bonsai-4b)
- [Ternary Bonsai 8B (Prism ML)](https://modelscale.dev/model-profile?model=ternary-bonsai-8b)
- [Thunder-LLM 8B (Academic)](https://modelscale.dev/model-profile?model=thunder-llm)
- [Toast 1 (Mixedbread)](https://modelscale.dev/model-profile?model=toast-1)
- [UI-Mate-27B (Tencent)](https://modelscale.dev/model-profile?model=ui-mate-27b)
- [UI-Mate-9B (Tencent)](https://modelscale.dev/model-profile?model=ui-mate-9b)
- [Ultravox GLM-4P6 (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-glm-4p6)
- [Ultravox GLM-4P7 (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-glm-4p7)
- [Ultravox v0.4.1 Llama 3.1 8B (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-v0-4-1-llama-3-1-8b)
- [Ultravox v0.5 Llama 3.1 8B (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-v0-5-llama-3-1-8b)
- [Ultravox v0.5 Llama 3.2 1B (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-v0-5-llama-3-2-1b)
- [Ultravox v0.6 Llama 3.3 70B (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-v0-6-llama-3-3-70b)
- [Ultravox v0.7 (Fixie AI)](https://modelscale.dev/model-profile?model=ultravox-v0-7)
- [Varco (NC AI)](https://modelscale.dev/model-profile?model=varco)
- [VibeVoice-ASR-Streaming 1.5B (Microsoft)](https://modelscale.dev/model-profile?model=vibevoice-asr-streaming-1-5b)
- [VibeVoice-ASR-Streaming 7B (Microsoft)](https://modelscale.dev/model-profile?model=vibevoice-asr-streaming-7b)
- [VITA 1.0 (VITA-MLLM)](https://modelscale.dev/model-profile?model=vita-1-0)
- [VITA 1.5 (VITA-MLLM)](https://modelscale.dev/model-profile?model=vita-1-5)
- [Voxtral 4B TTS 2603 (Mistral AI)](https://modelscale.dev/model-profile?model=voxtral-4b-tts-2603)
- [ZAYA1-74B-Preview (Zyphra)](https://modelscale.dev/model-profile?model=zaya1-74b-preview)
- [ZAYA1-8B (Zyphra)](https://modelscale.dev/model-profile?model=zaya1-8b)
- [ZAYA1-VL-8B (Zyphra)](https://modelscale.dev/model-profile?model=zaya1-vl-8b)
- [Zonos 2 (Zyphra)](https://modelscale.dev/model-profile?model=zonos-2)

## Articles

- [How to Track Claude Code Usage, Tokens, and Spend](https://modelscale.dev/articles/track-claude-code-usage): Claude Code can consume an individual Claude subscription, an eligible Team or Enterprise seat, or usage from an Anthropic Console API key. Those paths do not report cost in the…
- [How to Monitor OpenAI Codex Usage, Credits, and Token Costs](https://modelscale.dev/articles/monitor-openai-codex-usage): A Codex task launched through a ChatGPT entitlement or flexible credit balance is not the same billing stream as an application calling the OpenAI Platform API. The interfaces…
- [OpenRouter for Beginners: One API, Model Routing, and Cost Controls](https://modelscale.dev/articles/openrouter-guide-model-routing-cost-controls): OpenRouter offers one API surface for many models and providers. It is useful when an application needs model choice, provider fallbacks, centralized credits, or normalized…
- [5 Legitimate Free AI API Options—and the Limits That Matter](https://modelscale.dev/articles/legitimate-free-ai-api-access-credits): A free AI API may provide a recurring quota, selected zero-price models, a daily compute allowance, or rate-limited prototyping access. Those are different products. None should…
- [Lower LLM API Costs with Caching, Batch Jobs, and Output Caps](https://modelscale.dev/articles/reduce-llm-api-costs-caching-batch-output-limits): Start with a week of representative requests grouped by product workflow: support classification, document extraction, coding assistance, research, or content generation. Record…
- [A hybrid router for high-stakes agentic work](https://modelscale.dev/articles/hybrid-router): Begin with a narrow, reviewable split: route multi-file agentic coding and high-ambiguity tasks to the higher-capability lane; route stable classification, extraction, and…

## Sources

Every upstream fetch behind the figures above, and what each one supplies. https://modelscale.dev/sources shows when each was last read and how many records it returned.

- [benchlm:models](https://benchlm.ai/data/models.json): Which models exist, their identity, limits, and access mode.
- [benchlm:leaderboard](https://benchlm.ai/data/leaderboard.json): Published overall rank and the seven capability scores.
- [benchlm:benchmarks](https://benchlm.ai/data/benchmarks.json): Individual benchmark task results, grouped by category.
- [benchlm:pricing](https://benchlm.ai/data/pricing.json): List token prices: input, output, and cache rates where published.
- [benchlm:speed](https://benchlm.ai/data/speed.json): Measured time to first token and output throughput.
- [benchlm:comparisons](https://benchlm.ai/data/comparisons.json): Pairwise category deltas used by the compare page.
- [benchlm:lifecycle](https://benchlm.ai/api/data/ai-model-changes): Release and retirement notices with their effective dates.
- [livebench:table](https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv): Per-task LiveBench scores for the models it covers.
- [livebench:cost](https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/cost_2026_06_25.csv): The cost figures LiveBench publishes beside those scores.
- [openrouter:pricing](https://openrouter.ai/api/v1/models): Hosted token prices, used only where the primary source publishes none.
- [openrouter:endpoints](https://openrouter.ai/api/v1/models/{id}/endpoints): Per-endpoint throughput observations for hosted deployments.

## Optional

- [Sitemap](https://modelscale.dev/sitemap.xml): Every page, every model profile and every article, dated by the source's own generation time.
- [robots.txt](https://modelscale.dev/robots.txt): Crawl rules. Everything is allowed except /api, which serves this site's own JSON.