# GPT-6 Astra

OpenAI · Proprietary · rank 5 · bench-align-v5

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/models/gpt-6-astra  
JSON: https://modelscale.dev/api/model/gpt-6-astra

## Facts

| Field | Value | Evidence |
| --- | --- | --- |
| Canonical id | `gpt-6-astra` | — |
| Overall score | 83.2 | Observed 2026-11-21 · source benchlm:models |
| Verified score (BenchLM): a separate BenchLM figure, not the overall score | 82 | Observed 2026-11-21 · source benchlm:models |
| Context window | 1.05M tokens | Observed 2026-11-21 · source benchlm:models |
| Max output (models.dev) | 128,000 tokens | Observed 2026-11-21 · source modelsdev:catalog |
| Modalities (models.dev) | Input: text, image, PDF · Output: text | Observed 2026-11-21 · source modelsdev:catalog |
| Release date | 2026-09-03 | Observed 2026-11-21 · source benchlm:models |
| Access type | Proprietary | — |
| Blended $/1M (75% input / 25% output) | $20.00 | Derived from the input and output rates below |

## Capability evidence

Seven axes from the ranking source. An axis the source did not score is unavailable, not zero. Beside each score is BenchLM's separate verified score for that axis; the two are published side by side and are not interchangeable.

| Axis | Score | Evidence | Verified (BenchLM) | Verified evidence |
| --- | --- | --- | --- | --- |
| Agentic | 86.4 | Observed 2026-11-21 · source benchlm:models | 80 | Observed 2026-11-21 · source benchlm:models |
| Coding | 75 | Observed 2026-11-21 · source benchlm:models | Unavailable | Unavailable · source benchlm:models |
| Knowledge | 85.4 | Observed 2026-11-21 · source benchlm:models | 84.9 | Observed 2026-11-21 · source benchlm:models |
| Reasoning | 89.4 | Observed 2026-11-21 · source benchlm:models | 89.4 | Observed 2026-11-21 · source benchlm:models |
| Multimodal & Grounded | 87.8 | Observed 2026-11-21 · source benchlm:models | 87.8 | Observed 2026-11-21 · source benchlm:models |
| Instruction Following | Unavailable | Unavailable · source benchlm:models | Unavailable | Unavailable · source benchlm:models |
| Math | 93.4 | Observed 2026-11-21 · source benchlm:models | 93.4 | Observed 2026-11-21 · source benchlm:models |

## Runtime service evidence

Measured values with the date they were observed. Nothing is inferred from a sibling model or a provider claim. Regional or per-endpoint measurements appear only when the API supplies them; none are modelled.

| Measurement | Value | Observed | Last good | Evidence |
| --- | --- | --- | --- | --- |
| Time to first token | 204.16 s | 2026-11-21 | 2026-11-21 | Observed 2026-11-21 · source benchlm:speed |
| Throughput | 42 tok/s | 2026-11-21 | 2026-11-21 | Observed 2026-11-21 · source benchlm:speed |

## Endpoint and price matrix

Every published price component, including cache reads and writes.

| Component | USD | Evidence |
| --- | --- | --- |
| Input / 1M tokens (base rate) | $10.00 | Observed 2026-11-21 · source benchlm:pricing |
| Output / 1M tokens (base rate) | $50.00 | Observed 2026-11-21 · source benchlm:pricing |
| Cache read / 1M tokens (base rate) | $1.00 | Observed 2026-11-21 · source benchlm:pricing |
| Cache write / 1M tokens (base rate) | $12.50 | Observed 2026-11-21 · source openrouter:pricing |
| Blended / 1M (75% input / 25% output) | $20.00 | Derived — from the input and output rates above; it has no source record of its own |
| Cost per successful task (LiveBench) | Unavailable | Unavailable · source livebench:table |
| Batch input / 1M tokens (OpenRouter batch) (base rate) | $5.00 | Observed 2026-11-21 · source openrouter:pricing |
| Batch output / 1M tokens (OpenRouter batch) (base rate) | $25.00 | Observed 2026-11-21 · source openrouter:pricing |
| Batch cache read / 1M tokens (OpenRouter batch) (base rate) | $0.50 | Observed 2026-11-21 · source openrouter:pricing |
| Batch cache write / 1M tokens (OpenRouter batch) (base rate) | $6.25 | Observed 2026-11-21 · source openrouter:pricing |

Batch API rate (asynchronous jobs), from OpenRouter's batch listing.

### Long-context price tiers (models.dev)

The standard input, output and cache rates above are the base rate. models.dev publishes higher rates for requests whose prompt exceeds the thresholds below. The batch rates have their own long-context tiers, from OpenRouter's batch listing, shown separately. The cost simulator prices a standard-rate request whose whole prompt is above a threshold entirely at the highest of these tiers it exceeds, and a batch request only at a batch tier; it never applies these tiers to batch rates. The cost example on this page uses the base rate: its 1,200-token prompt is below every threshold.

| Prompt size | USD per 1M tokens | Evidence |
| --- | --- | --- |
| Above 272K prompt tokens | $20.00 input / $75.00 output per 1M · cache read $2.00 · cache write $25.00 | Observed 2026-11-21 · source modelsdev:catalog |

### Long-context batch price tiers (OpenRouter batch)

The batch rates above are the base batch rate. OpenRouter's batch listing, the same listing as those rates, publishes higher batch rates for requests whose total prompt is above the thresholds below. The cost simulator prices a batch request whose whole prompt is above a threshold entirely at the highest batch tier it exceeds.

| Prompt size | Batch USD per 1M tokens | Evidence |
| --- | --- | --- |
| Above 272K prompt tokens | $10.00 input / $37.50 output per 1M · cache read $1.00 · cache write $12.50 | Observed 2026-11-21 · source openrouter:pricing |

**Pricing note.** OpenAI's GPT-6 Astra model page lists $10.00 input / $1.00 cached input / $12.50 cache writes / $50.00 output per million tokens, a 1,050,000-token context window, 922,000 maximum input tokens, and 128,000 maximum output tokens. Prompts above 272K input tokens are charged at 2x input and cache rates and 1.5x output for the full request; Batch and Flex are 50% of Standard and Fast mode is 2x. The September 3, 2026 price sheet lists Astra at 2.5x GPT-5.6 Sol's current $4.00 / $20.00 rates.

## Workload-aware monthly cost example

10 conversations per day × 8 messages × 22 active days, 1200 input and 400 output tokens per message, no cache. Derived here from the published rates above by this site's own calculator — not a figure any source published.

| Field | Value |
| --- | --- |
| Modelled monthly cost | $56.32 |
| Modelled tokens | 2.82M |

## Benchmark record

132 matched benchmark rows with their published value, unit, and provenance.

### Knowledge

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| HealthBench (raw) (HealthBench raw score) | 56.9 | 5,000 multi-turn patient conversations | Realistic healthcare conversations | [HealthBench: Evaluating Large Language Models Towards Improved Human Health](https://arxiv.org/abs/2505.08775) |
| HealthBench (length-adjusted) (HealthBench length-adjusted score) | 58.3 | 5,000 multi-turn patient conversations | Realistic healthcare conversations | [HealthBench: Evaluating Large Language Models Towards Improved Human Health](https://arxiv.org/abs/2505.08775) |
| HealthBench Professional (raw) (HealthBench Professional raw score) | 68.2 | 525 physician-authored conversations | Professional clinical tasks | [HealthBench Professional: Evaluating Large Language Models on Real Clinician Chats](https://arxiv.org/abs/2604.27470) |
| GPQA (Graduate-Level Google-Proof Q&A) | 96 | 448 questions | Graduate level | [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://arxiv.org/abs/2311.12022) |
| GPQA-D (GPQA Diamond) | 96.0 | Graduate-level science questions | Graduate level | [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://arxiv.org/abs/2311.12022) |
| Artificial Analysis Intelligence Index | 56.9 | Cross-benchmark intelligence index | Display-only external reference | [Artificial Analysis](https://artificialanalysis.ai/) |
| AA-GPQA Diamond (Artificial Analysis GPQA Diamond) | 96.1 | Graduate-level science questions | Graduate-level science reasoning | [Artificial Analysis GPQA Diamond Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/gpqa-diamond) |
| AA-HLE (Artificial Analysis Humanity's Last Exam) | 54.7 | Expert-level questions | Frontier expert reasoning | [Artificial Analysis Humanity's Last Exam Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/hle) |
| AA-Omniscience Index (Artificial Analysis Omniscience Index) | 42.4 | Knowledge questions | Broad factual knowledge | [AA-Omniscience: Knowledge and Hallucination Benchmark](https://artificialanalysis.ai/evaluations/omniscience) |
| AA-Omniscience Accuracy (Artificial Analysis Omniscience Accuracy) | 61.6 | Knowledge questions | Broad knowledge | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| AA-Omniscience Hallucination Rate (Artificial Analysis Omniscience Hallucination Rate) | 49.8 | Knowledge questions | Factuality | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| HealthBench Hard | 36.6 | 1,000 health prompts | Advanced health reasoning | [Muse Spark Eval Methodology](https://ai.meta.com/static-resource/muse-spark-eval-methodology) |
| HealthBench Professional | 64.7 | Clinician chat tasks | Professional clinical workflows | [HealthBench Professional: Evaluating Large Language Models on Real Clinician Chats](https://arxiv.org/abs/2604.27470) |
| Morphology-to-molecule matching (Axiom Bio morphology-to-molecule matching) | 22.8 | Four multi-channel cell-imaging perturbation-matching tasks | Biomedical image and molecular reasoning | [Claude Opus 5.5 System Card](https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf) |
| HealthBench (raw, Artificial Analysis) (Artificial Analysis HealthBench raw score) | 56.9 | 5,000 conversations | Health conversation evaluation | [HealthBench independent evaluation](https://artificialanalysis.ai/evaluations/healthbench) |
| HealthBench (length-adjusted, Artificial Analysis) (Artificial Analysis HealthBench length-adjusted score) | 58.2 | 5,000 conversations | Health conversation evaluation | [HealthBench independent evaluation](https://artificialanalysis.ai/evaluations/healthbench) |
| HealthBench Professional (length-adjusted, Artificial Analysis) (Artificial Analysis HealthBench Professional length-adjusted score) | 66.0 | 525 conversations | Health conversation evaluation | [HealthBench Professional independent evaluation](https://artificialanalysis.ai/evaluations/healthbench-professional) |
| HealthBench Professional (raw, Artificial Analysis) (Artificial Analysis HealthBench Professional raw score) | 69.7 | 525 conversations | Health conversation evaluation | [HealthBench Professional independent evaluation](https://artificialanalysis.ai/evaluations/healthbench-professional) |
| PhysicianBench | 48.0 | 100 EHR physician tasks | Clinical agent work | [Claude Fable 5.2 and Claude Mythos 5.2 system card](https://assets.claude.com/27d60918d218125e7504bc7354c4b4626ec56b11.pdf) |

### Coding

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| FrontierCode 1.1 Main | 53.3 | 100 private Main tasks (150 in Extended) | Frontier coding-agent quality | [FrontierCode leaderboard](https://cognition.com/frontiercode) |
| FrontierCode 1.1 Extended | 64.5 | 150 private software-engineering tasks | Frontier coding-agent quality | [FrontierCode leaderboard](https://cognition.com/frontiercode) |
| FrontierSWE v2 | 65.5 | 34 ultra-long-horizon engineering and research tasks | Ultra-long-horizon frontier software engineering | [FrontierSWE v2](https://www.frontierswe.com/blog/v2) |
| Bug Hunt Bench | 45.0 | 105 planted bugs across two production TypeScript repositories | Blind production-repository bug finding and repair | [Bug Hunt Bench method, caveats, and definitions](https://bughunt.productcompass.pm/method) |
| AA Coding Index (Artificial Analysis Coding Index) | 76.9 | Cross-benchmark coding index | Display-only external reference | [Artificial Analysis model leaderboards](https://artificialanalysis.ai/leaderboards/models) |
| AA-SciCode (Artificial Analysis SciCode) | 56.5 | Scientific coding subproblems | Scientific programming | [Artificial Analysis SciCode Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/scicode) |
| DeepSWE | 74.1 | 113 software engineering tasks across 91 repositories and 5 languages | Long-horizon software engineering | [DeepSWE benchmark blog](https://deepswe.datacurve.ai/blog) |
| PostTrainBench v1.1 | 44.3 | Post-training Qwen3 1.7B and 4B, SmolLM3 3B, and Gemma3 4B | Frontier agent evaluation | [PostTrainBench v1.1 leaderboard and methodology](https://posttrainbench.com/?version=v1.1) |
| VulcanBench Frontier v4 | 89.3 | 23 behavioural-reconstruction coding tasks | Professional multi-file software engineering | [VulcanBench Frontier v4 leaderboard](https://vulcanbench.com/leaderboard.html) |
| FrontierCode 2.0 | 44.0 | 132 private software-engineering tasks | Frontier coding-agent quality | [FrontierCode 2.0](https://cognition.com/blog/frontier-code-2.0) |
| ScarfBench (ScarfBench: Cross-Framework Application Migration in Enterprise Java) | 65.4 | 204 framework migration tasks across 34 enterprise Java applications | Whole-application framework migration | [ScarfBench: A Benchmark for Cross-Framework Application Migration in Enterprise Java](https://arxiv.org/abs/2605.06754) |
| PostTrainBench (PostTrainBench v1.2) | 41.9 | Post-training four base models on six benchmarks | Autonomous ML engineering | [PostTrainBench](https://posttrainbench.com/) |

### Mathematics

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| FrontierMath v2 (Tiers 1-3) (FrontierMath v2 Tiers 1-3) | 93.684 | 285 private advanced mathematics problems | From olympiad-plus to early research mathematics | [FrontierMath v2 benchmark hub](https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2) |
| FrontierMath v2 (Tier 4) (FrontierMath v2 Tier 4) | 97.600 | 41 private extreme-difficulty mathematics problems | Research-level mathematics requiring hours or days of expert work | [FrontierMath Tier 4 v2 leaderboard](https://epoch.ai/benchmarks/frontiermath-tier-4-v2?view=graph&tab=leaderboard) |

### Reasoning

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| ARC-AGI-1 (ARC-AGI-1 Semi-Private Evaluation) | 97.50 | Semi-private ARC-AGI-1 evaluation set | Abstract visual reasoning | [ARC Prize leaderboard](https://arcprize.org/) |
| MRCR v2 256K-512K (OpenAI MRCR v2 8-needle 256K-512K) | 100.0 | 100 eight-needle retrieval examples | Very long-context retrieval | [Muse Spark 1.3 Evaluation Methodology](https://research.meta.ai/static/muse-spark-1-3-multimodal-evaluation-methodology) |
| MRCR v2 512K-1M (OpenAI MRCR v2 8-needle 512K-1M) | 96.3 | 100 eight-needle retrieval examples | Million-token retrieval | [Muse Spark 1.3 Evaluation Methodology](https://research.meta.ai/static/muse-spark-1-3-multimodal-evaluation-methodology) |
| ARC-AGI-2 (Abstraction and Reasoning Corpus for AGI v2) | 95 | Visual pattern completion and abstract reasoning | Expert-level — hardest public reasoning benchmark | [ARC-AGI-2: A Harder General Intelligence Benchmark](https://arcprize.org/arc-agi/2/) |
| ARC-AGI-3 (Abstraction and Reasoning Corpus for AGI v3) | 62.712802 | Interactive game-like tasks with hidden rules | Frontier agentic reasoning | [ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence](https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf) |
| GeneBench-Pro | 37.8 | 129 genomics statistical-analysis workflows | Long-horizon scientific reasoning | [GeneBench-Pro: Evaluating Multistage Statistical Reasoning](https://cdn.openai.com/pdf/21938268-21af-442f-af93-3b2249afb241/genebench-pro.pdf) |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 80.7 | Long-context reasoning tasks | Long-context reasoning | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| MLCR-AA (Medical Long Context Reasoning (MLCR-AA)) | 41.7 | Long, fragmented medical-record reasoning | Long-context medical reasoning | [Medical Long Context Reasoning (MLCR-AA)](https://artificialanalysis.ai/evaluations/mlcr-aa) |
| CritPt (Critical Physics Tasks) | 31.7 | Research-level physics questions | Research-level physics reasoning | [CritPt Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/critpt) |
| GraphWalks BFS 256K–1M | 71.8 | Breadth-first-search graph traversal on 200 problems with context lengths from 256K to 1M tokens. | Frontier agent evaluation | [Gemini 4 Argon evaluation methodology](https://deepmind.google/models/evals-methodology/gemini-4-argon) |
| STEMBench (STEMBench120: six-domain scientific research progress) | 2.81 | 120 research tasks, 20 in each of six STEM domains | Open research targets; no run has reached a full-target success | [STEMBench: six-domain model evaluation](https://www.ulam.ai/stembench) |
| ARC-AGI-3 Provider Adapter (ARC-AGI-3 Semi-Private: Provider Adapter harness) | 98.55 | — | — | [ARC Prize verified ARC-AGI-3 results](https://arcprize.org/leaderboard) |
| BioBasicsBench (BioBasicsBench: everyday wet-lab science) | 67.0 | 100 routine wet-lab analysis tasks (88 multimodal) across five domains | Routine wet-lab data interpretation | [Measuring how AI agents handle everyday wet-lab science](https://openai.com/index/measuring-how-ai-agents-handle-wet-lab-science/) |

### Agentic

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Terminal-Bench 4.0 | 57.90 | 66 professional computer-work tasks | Frontier autonomous knowledge work | [Terminal-Bench 4.0](https://www.tbench.ai/news/terminal-bench-4-0) |
| Terminal-Bench-Science 0.1 | 64.6 | 70 expert-curated scientific research workflows | Frontier scientific research workflows | [Terminal-Bench-Science 0.1](https://www.terminal-bench-science.ai/) |
| AA Briefcase (Artificial Analysis Briefcase) | 1569 | Professional knowledge-work tasks | Professional work | [Artificial Analysis Briefcase Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/aa-briefcase) |
| AA AutomationBench (Artificial Analysis AutomationBench) | 68.5 | Business-process automation tasks | Agentic automation | [Artificial Analysis AutomationBench Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/automationbench-aa) |
| AA ITBench (Artificial Analysis ITBench-AA) | 48.6 | IT incident-response tasks | Enterprise IT operations | [Artificial Analysis ITBench-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/itbench-aa) |
| AA Tau3 Banking (Artificial Analysis Tau3-Banking) | 41.4 | Banking tool-use workflows | Agentic banking workflows | [Artificial Analysis Tau3-Banking Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/tau3-banking) |
| BrowseComp | 91.5 | Research questions requiring browsing | Hard web research | [BrowseComp](https://openai.com/index/browsecomp/) |
| HLE w/ tools (Humanity's Last Exam with tools) | 57.2 | Expert questions with tool use | Frontier tool-augmented reasoning | [DeepSeek-V4 Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) |
| GDPval-AA | 1580 | Agentic real-world work tasks | Professional agentic workflows | [DeepSeek-V4 Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) |
| GDPval-AA (GDPval-AA normalized) | 53.7 | Economically valuable tasks | Professional agentic workflows | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| AA Agentic Index (Artificial Analysis Agentic Index) | 51.5 | Cross-benchmark agentic index | Display-only external reference | [Artificial Analysis model leaderboards](https://artificialanalysis.ai/leaderboards/models) |
| AA Terminal-Bench 4.0 (Artificial Analysis Terminal-Bench v4.0) | 59.1 | Terminal-based agent tasks | Agentic software engineering | [Artificial Analysis Terminal-Bench v4.0 Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/terminalbench-v4-0) |
| GDP.pdf (Artificial Analysis GDP.pdf) | 31.0 | Professional document-production tasks | Professional knowledge work | [GDP.pdf Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/gdp-pdf) |
| AA-AnalystAgent (Artificial Analysis AnalystAgent) | 51.2 | Spreadsheet and document analysis questions | Business and data analysis | [AA-AnalystAgent Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/aa-analyst-agent) |
| OSWorld 2.0 | 72.6 | 108 long-horizon computer-use workflows | Long-horizon professional workflows | [OSWorld2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks](https://arxiv.org/abs/2606.29537) |
| ExploitGym | 42.4 | 898 exploitation tasks | Advanced cybersecurity exploitation | [ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?](https://arxiv.org/abs/2605.11086) |
| AutomationBench | 41.4 | 600 public automation tasks | Long-horizon automation | [Kimi K3: Open Frontier Intelligence](https://www.kimi.com/blog/kimi-k3) |
| Agents' Last Exam | 59.3 | Agent tasks | Advanced agentic work | [DeepSeek V4 Flash 0731 update](https://api-docs.deepseek.com/zh-cn/updates/) |
| Terminal-Bench 2.1 (Vals) (Terminal-Bench 2.1, Vals AI run) | 87.3 | Difficult terminal tasks | Frontier agentic | [Vals AI Terminal-Bench 2.1, Vals AI run leaderboard](https://www.vals.ai/benchmarks/terminal-bench-2-1) |
| ApprenticeBench (ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job) | 68 | 100 vendor bills processed in sequence inside a simulated construction company | Long-horizon computer use with offline and online continual learning | [ApprenticeBench: a step change in AI's job readiness](https://neocognition.io/blog/apprentice-bench/) |
| CWE-bench v1 | 68.0 | Audit and patch held-out C/C++, Go, Java, TypeScript/JavaScript, Python, and Rust repositories | Frontier agent evaluation | [CWE-bench v1 leaderboard and methodology](https://cwe-bench.com/) |
| Harvey LAB-AA v1.1 | 8.6 | 120 private legal tasks across 24 practice areas | Professional legal work | [Harvey LAB-AA v1.1 Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/harvey-lab-aa) |
| Hermes Index (Hermes Index: agentic capability and cost per task inside Hermes Agent) | 56.3 | Four suites: Hermes Bench (150 tasks), TerminalBench 4 without its 4 GPU tasks, TerminalBench Science and SkillsBench | Frontier agentic | [Hermes Index: an agentic model leaderboard](https://portal.nousresearch.com/bench) |
| Hermes Bench (Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behind) | 73.7 | 150 tasks across 25 categories | Frontier agentic | [Hermes Bench (Hermes Index)](https://portal.nousresearch.com/bench) |
| MazeBench (MazeBench: visual spatial reasoning in a 3D open world) | 24 | 100 hidden gems across 256 rooms in one open world | Long-horizon spatial planning, with single puzzles that can take over one hundred moves | [MazeBench Results](https://mazebench.com/blog?post=maze-bench-results) |
| AA-Briefcase (rubric) (AA-Briefcase v1.1, rubric pass rate) | 52.0 | Professional knowledge-work briefs | Professional knowledge work | [Artificial Analysis AA-Briefcase](https://artificialanalysis.ai/) |
| Remote Labor Index (Remote Labor Index (RLI)) | 20.8 | 240 freelance projects | End-to-end remote work | [Remote Labor Index](https://scale.com/leaderboard/rli) |
| RoboDojo Sim | 29.8 | Simulated robot-control tasks (count not published) | Embodied agent control | [Introducing Claude Fable 5.2 and Claude Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2) |
| CWE-bench (CWE-bench v2) | 64.2 | Audit and patch C/C++, Go, Java, TypeScript/JavaScript, Python, and Rust repositories; 100 public and 150 held-out tasks | Frontier agent evaluation | [CWE-bench v2 leaderboard and methodology](https://cwe-bench.com/) |
| Terminal-Bench 4.1 | 60.30 | 66 professional computer-work tasks | Frontier autonomous knowledge work | [Terminal-Bench 4.1](https://www.tbench.ai/news/terminal-bench-4-1) |
| OSWorld 2.1 (OSWorld 2.1 (all 108 tasks)) | 75.8 | 108 long-horizon computer-use workflows | Long-horizon professional workflows | [OSWorld 2.1 release (xlang-ai/OSWorld-V2, tag osworld-v2.1)](https://github.com/xlang-ai/OSWorld-V2/releases) |
| AA Cyber Index (Artificial Analysis Cyber Index) | 33.4 | Three defensive cybersecurity evaluations | Real-world vulnerability discovery and patching | [Artificial Analysis Cyber Index](https://artificialanalysis.ai/evaluations/artificial-analysis-cyber-index) |
| CWE-Bench-AA (Artificial Analysis CWE-Bench-AA) | 63.3 | Real-world vulnerability patching tasks | Automated vulnerability remediation | [CWE-Bench-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/cwe-bench-aa) |
| DeepsecBench-AA (Artificial Analysis DeepsecBench-AA) | 36.9 | Vulnerability discovery in real codebases | Security code review | [DeepsecBench-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/deepsecbench-aa) |
| CyberGym-E2E-AA (Artificial Analysis CyberGym-E2E-AA) | 0.0 | 131 end-to-end vulnerability reproduction and patching tasks | Real-world cybersecurity | [CyberGym-E2E-AA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/cybergym-e2e-aa) |
| MKA (Multimodal Knowledge Agent Benchmark) | 50.2 | 200 multimodal professional tasks across 10 industry verticals | Expert multimodal knowledge work | [MKA Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/mka) |
| LifeVal (LifeVal: everyday life tasks) | 80.6 | 169 everyday life tasks (caregiving, health, education, household administration, personal finance, shopping, government services) | Agentic personal-assistant work | [LifeVal](https://openai.com/index/lifeval/) |

### Multimodal & Grounded

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| BenchCAD Vision2Code (tools) (BenchCAD Vision2Code voxel IoU with tools) | 0.959 | 1,000-file Vision2Code subset | Programmatic CAD generation | [BenchCAD: A comprehensive, industry-standard benchmark for programmatic CAD](https://arxiv.org/abs/2605.10865) |
| AA-MMMU-Pro (Artificial Analysis MMMU-Pro) | 86.9 | Multimodal academic reasoning | Frontier multimodal | [Artificial Analysis MMMU-Pro Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/mmmu-pro) |
| ScreenSpot Pro | 92.7 | 1,581 grounding instructions | Professional GUI grounding | [ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use](https://arxiv.org/abs/2504.07981) |
| Biomedical image analysis (Anthropic biomedical-image-analysis evaluation) | 77.5 | 91 cases across 31 biomedical imaging tasks | Biomedical image analysis | [Claude Opus 5.5 System Card](https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf) |
| ParseBench-1 (ParseBench-1: document parsing for AI agents (LlamaIndex leaderboard, first release)) | 69.81 | 2,078 human-verified enterprise pages from insurance, finance and government documents | No method is strong on all five dimensions; charts and visual grounding separate the field | [ParseBench: A Document Parsing Benchmark for AI Agents](https://arxiv.org/abs/2604.08538) |
| ExtractBench (LlamaIndex) (ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)) | 91.91 | 370 documents, 4,869 pages, 8 business domains, 67 document types, each with a JSON Schema | Short documents are near solved; long record lists and grounding separate systems | [ExtractBench: A Benchmark for Schema-Guided Enterprise Document Extraction](https://arxiv.org/abs/2607.29677) |
| ExtractBench Pro (ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)) | 58 | 100 synthetic documents, 2,622 pages, 86 document types, six challenge groups | The best system completes 58 of 100 documents perfectly; several complete none | [ExtractBench Pro leaderboard and challenge descriptions](https://www.extractbench.ai/pro) |

### external

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| SRE-Bench (SRE-Bench binary reverse engineering) | 88.0 | 262 binary reverse-engineering instances | Binary reverse engineering | [GPT-6 Astra System Card](https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf) |
| ExploitBench (ExploitBench v8-bench) | 100 | V8 exploit synthesis runs | Browser exploitation and cybersecurity | [ExploitBench](https://exploitbench.ai/) |
| ACCR standard (Advanced Cyber Completion Rate — Standard Access) | 3.5 | Internal advanced-cyber request set | Cyber access and safeguards | [Expanding Daybreak as the cyber defense window narrows](https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/) |
| ACCR Daybreak Blue (Advanced Cyber Completion Rate — Daybreak Blue) | 3.5 | Internal advanced-cyber request set | Cyber access and safeguards | [Expanding Daybreak as the cyber defense window narrows](https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/) |
| SEC-Bench Pro | 85.4 | Security engineering tasks | Advanced cybersecurity | [Introducing GPT-5.6](https://openai.com/index/gpt-5-6/) |
| FrontierCyber | 38.1 | 197 dynamic cyber tasks | Easy through elite cyber operations | [FrontierCyber](https://www.irregular.com/research/frontiercyber) |
| CyScenarioBench success (CyScenarioBench Average Success Rate) | 59 | 11 cyber scenarios | Long-horizon cybersecurity | [Assessing GPT-5.6 Sol](https://www.irregular.com/research/assessing-gpt-5.6-sol) |
| CyScenarioBench solved (CyScenarioBench Scenarios Ever Solved) | 9 | 11 cyber scenarios | Long-horizon cybersecurity | [Assessing GPT-5.6 Sol](https://www.irregular.com/research/assessing-gpt-5.6-sol) |
| Atomic network attacks (Atomic Network Attack Simulation) | 100 | Atomic cyber tasks | Network attack simulation | [Assessing GPT-5.6 Sol](https://www.irregular.com/research/assessing-gpt-5.6-sol) |
| Atomic vulnerability research (Atomic Vulnerability Research and Exploitation) | 100 | Atomic cyber tasks | Vulnerability research and exploitation | [Assessing GPT-5.6 Sol](https://www.irregular.com/research/assessing-gpt-5.6-sol) |
| Atomic evasion (Atomic Evasion) | 52 | Atomic cyber tasks | Cybersecurity evasion | [Assessing GPT-5.6 Sol](https://www.irregular.com/research/assessing-gpt-5.6-sol) |
| MaxxBench (MaxxBench (mutated DeepSWE tasks)) | 71.9 | 89 mutated DeepSWE feature-implementation tasks, two attempts per model | Long-horizon software engineering under changed requirements | [Why MaxxBench (unLAP research post)](https://maxxbench.unlap.ai/research/why-maxxbench) |
| SpireClimb (SpireClimb (Slay the Spire 1 and 2)) | 34.0 | 60 Slay the Spire 1 runs (Ascension 20, Heart kill) and 75 Slay the Spire 2 runs (Ascension 10) per model | Maximum game difficulty, long-horizon computer use | [SpireClimb (Epoch AI)](https://epoch.ai/benchmarks/spireclimb) |
| Chess Puzzles (Chess Puzzles (Epoch AI)) | 74.0 | 100 programmatically generated chess positions with a single best move | Novel chess positions, single best move | [Chess Puzzles (Epoch AI)](https://epoch.ai/benchmarks/chess-puzzles) |
| Mystery Game Puzzles (Mystery Game Puzzles (Epoch AI)) | 84.0 | 100 single-best-move positions from an undisclosed game | Unfamiliar game positions, single best move | [Mystery Game Puzzles (Epoch AI)](https://epoch.ai/benchmarks/mystery-game-puzzles) |
| Furniture Assembly (Furniture Assembly Benchmark (Epoch AI)) | 80.0 | 60 photos of three IKEA builds, some with deliberate assembly mistakes | Spatial reasoning from technical diagrams to real photos | [Furniture Assembly Benchmark (Epoch AI)](https://epoch.ai/benchmarks/furniture-assembly) |
| EBR-bench (Earthborne Rangers (EBR-bench)) | 76.2 | Ten playthroughs of an Earthborne Rangers campaign segment per sample (eight learning, two scored) | Learning an unfamiliar game over multi-hour playthroughs | [Earthborne Rangers (EBR-bench), Epoch AI](https://epoch.ai/benchmarks/ebr-bench) |
| MirrorCode (MirrorCode (Epoch AI and METR)) | 60.0 | 30 reimplementation tasks (15 programs in two languages), three attempts each | Days-long software engineering without source code | [MirrorCode (Epoch AI)](https://epoch.ai/benchmarks/mirrorcode) |
| Epoch Automation Reports | 55.5 | 11 open-ended work tasks in five categories from Epoch AI's own work | Open-ended professional research and design work | [Epoch Automation Reports](https://epoch.ai/benchmarks/epoch-automation-reports) |
| Clocktower Bench (Clocktower Bench (Blood on the Clocktower)) | 1859 | 386 rated head-to-head games of Blood on the Clocktower (10 to 12 players) | Social deduction, deception and multi-agent reasoning | [Clocktower Bench leaderboard](https://clocktowerbench.com/) |
| Web Pentesting (black-box) (OpenAI Web Pentesting, black-box score) | 32.0 | Black-box web application penetration tests | Web vulnerability discovery | [GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents](https://openai.com/index/introducing-gpt-6-cyber/) |
| ExploitBench Internal Port (ExploitBench — Internal Port (June–August 2026)) | 39.0 | Recently disclosed V8 and Chrome vulnerabilities (June–August 2026) | Browser exploitation | [GPT-6 Cyber System Card](https://deploymentsafety.openai.com/gpt-6-cyber) |
| ExploitGym Hardened (ExploitGym Hardened (stronger exploit mitigations)) | 29.7 | ExploitGym tasks with stronger exploit mitigations | Exploitation against mitigations | [ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?](https://arxiv.org/abs/2605.11086) |
| ExploitGym Honeypot (ExploitGym Honeypot out-of-scope exploitation rate) | 0.00 | Hard ExploitGym tasks with an out-of-scope honeypot service | Cyber-agent scope adherence | [GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents](https://openai.com/index/introducing-gpt-6-cyber/) |
| FACTS Multimodal | 55.2 | About 1,500 image questions: 711 public and 811 held out | Image-grounded factual answering | [BenchLM](https://benchlm.ai/benchmarks/facts-multimodal) |
| DAYJOB | 17.4 | 130 long-horizon tasks: 80 finance and 50 healthcare | Long-horizon professional work | [BenchLM](https://benchlm.ai/benchmarks/dayjob) |
| DAYJOB Healthcare | 11.2 | 50 long-horizon healthcare tasks across five subfields | Long-horizon clinical and healthcare operations work | [BenchLM](https://benchlm.ai/benchmarks/dayjob-healthcare) |
| FailureSensorIQ | 69.6 | Multiple-choice questions on sensors and failure modes for 10 industrial assets | Industrial maintenance and reliability knowledge | [BenchLM](https://benchlm.ai/benchmarks/failuresensoriq) |
| Code2Video | 1565 | Creative briefs rendered as motion-graphics video | Code-driven motion design | [BenchLM](https://benchlm.ai/benchmarks/code2video) |
| Landing Page Arena | 1657 | 12 client briefs for small-business landing pages, rated by human votes | Marketing page design and copy | [BenchLM](https://benchlm.ai/benchmarks/landing-page-arena) |
| Chart Arena | 1621 | 13 chart briefs with small datasets, rated by human votes | Data visualisation | [BenchLM](https://benchlm.ai/benchmarks/chart-arena) |
| Joke Arena | 1568 | 100 joke prompts in 10 categories, rated by human votes | Humour writing | [BenchLM](https://benchlm.ai/benchmarks/joke-arena) |
| Pelican Arena | 1817 | One prompt: an animated SVG of a pelican riding a bicycle, rated by human votes | Vector drawing and animation in code | [BenchLM](https://benchlm.ai/benchmarks/pelican-arena) |
| Landmark Arena | 1751 | 20 famous landmarks built as three.js web pages, rated by human votes | 3D scenes in browser code | [BenchLM](https://benchlm.ai/benchmarks/landmark-arena) |
| ASCII Arena | 1731 | ASCII drawings of subjects from Google's Quick, Draw! dataset, rated by human votes | Text-character drawing | [BenchLM](https://benchlm.ai/benchmarks/ascii-arena) |
| Game Arena Go | 1735 | Head-to-head Go games on a 13 x 13 board | Strategic planning and rule-following | [BenchLM](https://benchlm.ai/benchmarks/game-arena-go) |
| BabaIsBench (BabaIsBench (Baba Is You intro levels, pass@1)) | 100 | 8 intro levels of Baba Is You, three attempts per level (Stage 0); 15 further levels run once (Stage 1) | Rule-rewriting puzzles that need planning over many moves | [Baba Is Solved by Fable 5 and GPT-5.6 Sol, but at what cost? (Quesma)](https://quesma.com/blog/baba-is-bench/) |
| NYT Connections (NYT Connections LLM Benchmark (extended version)) | 98.4 | 940 NYT Connections puzzles, each with up to four added trick words | Wordplay, word knowledge and decoys that fit almost-but-not-quite | [NYT Connections LLM Benchmark (Lech Mazur, GitHub)](https://github.com/lechmazur/nyt-connections) |
| StealthMark (StealthMark Intelligence Benchmarks (overall score)) | 74.6 | 8 original tasks: Newton's Mirror, Ground Truth, Latch, Foresight, Mona Lisa, Selfie, Beach and Raccoon | Chained physical reasoning, planning without tools and precise visual construction | [StealthMark: The AI Benchmark Explained (Stealth Models)](https://stealthmodels.com/stealthmark) |
| FrontierPhysics (End-to-End Physics Research Benchmark for AI Agents) | 23.2 | 56 research projects from 42 physicists across nine fields, three trials each (168 rollouts per setup) | Open-ended research projects that took their authors about 270 days on average | [FrontierPhysics: Evaluating Agents for End-to-End Physics Research (arXiv, November 4, 2026)](https://arxiv.org/abs/2611.06910) |
| HoneyBench (HoneyBench v0.2 (Goodhart Labs)) | 50.8 | 20 private honeypot tasks across 7 domains with exploitable evaluators | Agentic alignment and specification gaming | [HoneyBench v0.2](https://goodhartlabs.com/honeybench) |

### Instruction Following

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Gray Swan IPI (15 attempts) (Gray Swan Indirect Prompt Injection, attack success at 15 attempts) | 8.5 | Indirect prompt-injection attacks | Adversarial prompt injection | [Gemini 4 Argon evaluation methodology](https://deepmind.google/models/evals-methodology/gemini-4-argon) |

## Independent scores (Epoch AI)

A second, independent score source: Epoch's own figures in Epoch's units, one line per model version Epoch evaluated (effort setting as Epoch labels it); higher is better. Not part of this site's capability scores or rank.

| Benchmark | Variant | Score | Unit | Epoch date | Evidence |
| --- | --- | --- | --- | --- | --- |
| GPQA Diamond | max | 0.9577 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tiers 1-3 (v2) | max | 0.9368 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | high | 0.976 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | low | 0.878 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | max | 0.976 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | medium | 0.9756 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | none | 0.8293 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| FrontierMath Tier 4 (v2) | xhigh | 0.976 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |
| SimpleQA Verified | max | 0.756 | fraction (0–1) | 2026-08-30 | Last good 2026-11-20 · source epoch:benchmarks |

Source: [Epoch AI, CC BY 4.0](https://epoch.ai/benchmarks) ([licence](https://creativecommons.org/licenses/by/4.0/)). Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from https://epoch.ai/benchmarks.

## Lifecycle and limitations log

Lifecycle events the source associates with this model.

| Event | Type | Confirmation | Announced | Effective | Replacement | Source |
| --- | --- | --- | --- | --- | --- | --- |
| GPT-6 Astra | Release | confirmed | Unavailable | 2026-09-03 | — | [OpenAI GPT-6 Astra model documentation](https://developers.openai.com/api/docs/models/gpt-6-astra) |

## What this profile does not claim

Values are reproduced exactly as their sources published them, in the units those sources declared; none are converted, interpolated, or averaged across providers. Any field marked unavailable was attempted and not returned. Last attempted fetch for this model's score: 2026-11-21 01:17 UTC.
