# MazeBench leaderboard

Agentic benchmark

> 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/leaderboards/benchmarks/agentic-maze-bench

MazeBench: visual spatial reasoning in a 3D open world. Every model the catalog carries a published MazeBench value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Agentic |
| Measure | Gems collected out of 100, with rooms visited as a second score |
| Tasks | 100 hidden gems across 256 rooms in one open world |
| Difficulty | Long-horizon spatial planning, with single puzzles that can take over one hundred moves |
| Published by | [MazeBench Results](https://mazebench.com/blog?post=maze-bench-results) |

Tests whether a coding agent can explore a 256-room 3D puzzle world and collect its 100 hidden gems, using box-pushing puzzles, camera rotation, and long-horizon planning.

## MazeBench ranking

43 models with a published MazeBench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Gems collected out of 100, with rooms visited as a second score | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 23 | Observed 2026-10-23 · source benchlm:models |
| 2 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 17 | Observed 2026-10-23 · source benchlm:models |
| 3 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 13 | Observed 2026-10-23 · source benchlm:models |
| 3 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 13 | Observed 2026-10-23 · source benchlm:models |
| 5 | [Claude Opus 5](https://modelscale.dev/models/claude-opus-5) | Anthropic | 12 | Observed 2026-10-23 · source benchlm:models |
| 5 | [Gemini 4 Argon](https://modelscale.dev/models/gemini-4-argon) | Google | 12 | Observed 2026-10-23 · source benchlm:models |
| 7 | [Claude Fable 5](https://modelscale.dev/models/claude-fable-5) | Anthropic | 11 | Observed 2026-10-23 · source benchlm:models |
| 8 | [Claude Fable 5.1](https://modelscale.dev/models/claude-fable-5-1) | Anthropic | 10 | Observed 2026-10-23 · source benchlm:models |
| 9 | [Claude Sonnet 5.5](https://modelscale.dev/models/claude-sonnet-5-5) | Anthropic | 8 | Observed 2026-10-23 · source benchlm:models |
| 10 | [GPT-6 Sol](https://modelscale.dev/models/gpt-6-sol) | OpenAI | 7 | Observed 2026-10-23 · source benchlm:models |
| 11 | [GPT-5.6 Terra](https://modelscale.dev/models/gpt-5-6-terra) | OpenAI | 6 | Observed 2026-10-23 · source benchlm:models |
| 12 | [Gemini 3.8 Flash](https://modelscale.dev/models/gemini-3-8-flash) | Google | 4 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Claude Haiku 5.5](https://modelscale.dev/models/claude-haiku-5-5) | Anthropic | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Gemini 3.1 Pro](https://modelscale.dev/models/gemini-3-1-pro) | Google | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Gemini 3.7 Flash](https://modelscale.dev/models/gemini-3-7-flash) | Google | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Grok 4.6](https://modelscale.dev/models/grok-4-6) | xAI | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Grok 4.7](https://modelscale.dev/models/grok-4-7) | xAI | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Kimi K3](https://modelscale.dev/models/kimi-3) | Moonshot AI | 1 | Observed 2026-10-23 · source benchlm:models |
| 13 | [Muse Wonder Full](https://modelscale.dev/models/muse-wonder-full) | Meta | 1 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Claude Haiku 4.5](https://modelscale.dev/models/claude-haiku-4-5) | Anthropic | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Claude Opus 4.8](https://modelscale.dev/models/claude-opus-4-8) | Anthropic | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Claude Sonnet 4.6](https://modelscale.dev/models/claude-sonnet-4-6) | Anthropic | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Claude Sonnet 5](https://modelscale.dev/models/claude-sonnet-5) | Anthropic | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [DeepSeek V4 Flash 0731](https://modelscale.dev/models/deepseek-v4-flash-max) | DeepSeek | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Gemini 3.5 Flash](https://modelscale.dev/models/gemini-3-5-flash) | Google | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Gemini 3.6 Flash](https://modelscale.dev/models/gemini-3-6-flash) | Google | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GLM-5.2](https://modelscale.dev/models/glm-5-2) | Z.AI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GLM-5.3](https://modelscale.dev/models/glm-5-3) | Z.AI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GLM-5.3-Flash](https://modelscale.dev/models/glm-5-3-flash) | Z.AI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-5.3-Codex-Spark](https://modelscale.dev/models/gpt-5-3-codex-spark) | OpenAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-5.4](https://modelscale.dev/models/gpt-5-4) | OpenAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-5.5](https://modelscale.dev/models/gpt-5-5) | OpenAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-5.6 Luna](https://modelscale.dev/models/gpt-5-6-luna) | OpenAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-6 Luna](https://modelscale.dev/models/gpt-6-luna) | OpenAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Grok 4.20](https://modelscale.dev/models/grok-4-20-beta) | xAI | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Llama 4 Maverick](https://modelscale.dev/models/llama-4-maverick) | Meta | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [MiMo-V2.5-Pro](https://modelscale.dev/models/mimo-v2-5-pro) | Xiaomi | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [MiniMax M3](https://modelscale.dev/models/minimax-m3) | MiniMax | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Muse Spark 1.2](https://modelscale.dev/models/muse-spark-1-2) | Meta | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Muse Spark 1.3](https://modelscale.dev/models/muse-spark-1-3) | Meta | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Muse Wonder](https://modelscale.dev/models/muse-wonder) | Meta | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Nemotron 3 Ultra](https://modelscale.dev/models/nemotron-3-ultra-500b) | NVIDIA | 0 | Observed 2026-10-23 · source benchlm:models |
| 20 | [Qwen3.8 Max](https://modelscale.dev/models/qwen3-8-max) | Alibaba | 0 | Observed 2026-10-23 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.