MaxxBench
Agents solve 89 modified DeepSWE repository tasks. Revised requirements, code, reference patches, and tests check whether agents can solve changed problems rather than recall original fixes.
The maker dates the methodology October 20. Task success requires passing the modified tests.
- Made by
- unLAP
- Released
- October 2026
- Kind
- Fixing real repositories
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Items from
- Another benchmark, Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- DeepSWE
- In views
- Agents
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- October 2026
- Scores refreshed
- November 18th, 2026
Claude Opus 5.5 leads with 76.4%
Best score per model from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
- 76.4 Claude Opus 5.5
- 71.9 GPT-6 Astra
- 69.1 GPT-5.6 Sol
- 66.9 Claude Sonnet 5.5
- 63.5 Claude Fable 5
- 63.5 Claude Opus 5
- 59.8 Claude Haiku 5.5
- 59.0
DeepSeek V4.1 Flash 1017
- 56.7 GLM-5.3
- 55.6
DeepSeek V4.1 Flash
- 55.1 GLM-5.3-Flash
- 53.4
Gemini 3.8 Flash
- 51.7 GPT-6 Luna
- 46.1 GPT-5.6 Luna
- 46.1
Qwen3.8-Flash-Next
- 38.8
Muse Spark 1.2
- 28.1
DeepSeek-V4-Flash
- 1 Claude Opus 5.5 max 76.4%
- 2 GPT-6 Astra xhigh 71.9%
- 3 GPT-5.6 Sol max 69.1%
- 4 Claude Sonnet 5.5 max 66.9%
- 5 Claude Fable 5 xhigh 63.5%
- 6 Claude Opus 5 max 63.5%
- 7 Claude Haiku 5.5 max 59.8%
- 8
DeepSeek V4.1 Flash 1017 max 59%
- 9 GLM-5.3 max 56.7%
- 10
DeepSeek V4.1 Flash max 55.6%
- 11 GLM-5.3-Flash max 55.1%
- 12
Gemini 3.8 Flash high 53.4%
- 13 GPT-6 Luna max 51.7%
- 14 GPT-5.6 Luna max 46.1%
- 15
Qwen3.8-Flash-Next xhigh 46.1%
- 16
Muse Spark 1.2 xhigh 38.8%
- 17
DeepSeek-V4-Flash max 28.1%
Source: Official MaxxBench leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 76.4% | max | Sep 2026 | Official leaderboard |
| 2 | GPT-6 Astra | 71.9% | xhigh | Sep 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 69.1% | max | Jul 2026 | Official leaderboard |
| 4 | Claude Sonnet 5.5 | 66.9% | max | Sep 2026 | Official leaderboard |
| 5 | Claude Fable 5 | 63.5% | xhigh | Jun 2026 | Official leaderboard |
| 6 | Claude Opus 5 | 63.5% | max | Jul 2026 | Official leaderboard |
| 7 | Claude Haiku 5.5 | 59.8% | max | Oct 2026 | Official leaderboard |
| 8 | | 59% | max | Oct 2026 | Official leaderboard |
| 9 | GLM-5.3 | 56.7% | max | Aug 2026 | Official leaderboard |
| 10 | | 55.6% | max | Sep 2026 | Official leaderboard |
| 11 | GLM-5.3-Flash | 55.1% | max | Aug 2026 | Official leaderboard |
| 12 | | 53.4% | high | Sep 2026 | Official leaderboard |
| 13 | GPT-6 Luna | 51.7% | max | Sep 2026 | Official leaderboard |
| 14 | GPT-5.6 Luna | 46.1% | max | Jul 2026 | Official leaderboard |
| 15 | | 46.1% | xhigh | Aug 2026 | Official leaderboard |
| 16 | | 38.8% | xhigh | Aug 2026 | Official leaderboard |
| 17 | | 28.1% | max | Apr 2026 | Official leaderboard |
The best score, 76.4%, costs $16 a task
Runs on the cost frontier score higher than any run that's cheaper. GPT-6 Astra at xhigh comes within 4.5 points of the best for $4.38 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official MaxxBench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5.5 | max | 76.4% | $16 | ||
| GPT-6 Astra | xhigh | 71.9% | $4.38 | ||
| GPT-5.6 Sol | max | 69.1% | $5.95 | ||
| Claude Sonnet 5.5 | max | 66.9% | $11 | ||
| Claude Fable 5 | xhigh | 63.5% | $15 | ||
| Claude Opus 5 | max | 63.5% | $14 | ||
| Claude Haiku 5.5 | max | 59.8% | $2.83 | ||
| max | 59% | $0.26 | |||
| GLM-5.3 | max | 56.7% | $5.56 | ||
| max | 55.6% | $0.62 | |||
| GLM-5.3-Flash | max | 55.1% | $0.73 | ||
| high | 53.4% | $2.53 | |||
| GPT-6 Luna | max | 51.7% | $0.26 | ||
| GPT-5.6 Luna | max | 46.1% | $0.53 | ||
| xhigh | 46.1% | $0.36 | |||
| xhigh | 38.8% | $3.03 | |||
| max | 28.1% | $0.58 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official MaxxBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MaxxBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Coding
- SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
- LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests.
DeepSeek-V4-Pro 93.5%
- SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
- DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
- SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases.
Seed2.1 Pro 59.8%
- Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance.
DeepSeek V4.1 Flash 3,471
Data audit
All 11 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 5 pass
- Pass Independent runs Scores from the official leaderboard (17 entries).
- Pass Models scored 17 models we track.
- Pass Newest score The newest is from October 2026, for DeepSeek V4.1 Flash 1017.
- Pass Frontier models 3 of the top 10 have a score, led by GPT-6 Astra (#4).
- Pass Headroom The best is 76.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on maxxbench.unlap.ai: 17 rows for models we track at the last read.
- Pass Last read November 18th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.