# MaxxBench leaderboard

external 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/external-maxx-bench

MaxxBench (mutated DeepSWE tasks). Every model the catalog carries a published MaxxBench value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | external |
| Measure | Pass rate over 178 agent attempts, mini-swe-agent harness |
| Tasks | 89 mutated DeepSWE feature-implementation tasks, two attempts per model |
| Difficulty | Long-horizon software engineering under changed requirements |
| Published by | [Why MaxxBench (unLAP research post)](https://maxxbench.unlap.ai/research/why-maxxbench) |

unLAP's MaxxBench takes 89 DeepSWE software engineering tasks and changes the instructions or starting code while keeping the core problem, to test whether coding agents adapt to new requirements or fall back on a memorized solution.

## MaxxBench ranking

14 models with a published MaxxBench 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 | Pass rate over 178 agent attempts, mini-swe-agent harness | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 76.4 | Observed 2026-11-21 · source benchlm:models |
| 2 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 71.9 | Observed 2026-11-21 · source benchlm:models |
| 3 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 69.1 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Claude Sonnet 5.5](https://modelscale.dev/models/claude-sonnet-5-5) | Anthropic | 66.9 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Claude Fable 5](https://modelscale.dev/models/claude-fable-5) | Anthropic | 63.5 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Claude Opus 5](https://modelscale.dev/models/claude-opus-5) | Anthropic | 63.5 | Observed 2026-11-21 · source benchlm:models |
| 7 | [GLM-5.3](https://modelscale.dev/models/glm-5-3) | Z.AI | 56.7 | Observed 2026-11-21 · source benchlm:models |
| 8 | [DeepSeek V4.1 Flash 1017](https://modelscale.dev/models/deepseek-v4-1-flash) | DeepSeek | 55.6 | Observed 2026-11-21 · source benchlm:models |
| 9 | [GLM-5.3-Flash](https://modelscale.dev/models/glm-5-3-flash) | Z.AI | 55.1 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Gemini 3.8 Flash](https://modelscale.dev/models/gemini-3-8-flash) | Google | 53.4 | Observed 2026-11-21 · source benchlm:models |
| 11 | [GPT-6 Luna](https://modelscale.dev/models/gpt-6-luna) | OpenAI | 51.7 | Observed 2026-11-21 · source benchlm:models |
| 12 | [GPT-5.6 Luna](https://modelscale.dev/models/gpt-5-6-luna) | OpenAI | 46.1 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Muse Spark 1.2](https://modelscale.dev/models/muse-spark-1-2) | Meta | 38.8 | Observed 2026-11-21 · source benchlm:models |
| 14 | [DeepSeek V4 Flash 0731](https://modelscale.dev/models/deepseek-v4-flash-max) | DeepSeek | 28.1 | Observed 2026-11-21 · 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.
