HarnessOpt-Bench
Coding agents edit a target agent’s harness under a fixed evaluation budget. The grader measures improvement over the starting harness on held-out tasks.
The authors compared five models across four downstream tasks and 111 scored runs.
- Made by
- Scale AI
- Released
- August 2026
- Kind
- Machine learning engineering
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- Claude Opus 5 #11 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5 leads with 0.47 among models released by July 2026
Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall score, so this is the plain mean of its four task gains. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 13 more
- 1 Claude Opus 5 0.47
- 2 Kimi K3 0.32
- 3 GPT-5.6 Sol 0.28
- 4 Claude Sonnet 5 0.27
- 5 GPT-5.6 Terra 0.09
Source: The HarnessOpt-Bench paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 0.47 | opencode | Jul 2026 | Official leaderboard |
| 2 | Kimi K3 | 0.32 | kimi-cli | Jul 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 0.28 | codex | Jul 2026 | Official leaderboard |
| 4 | Claude Sonnet 5 | 0.27 | opencode | Jun 2026 | Official leaderboard |
| 5 | GPT-5.6 Terra | 0.09 | opencode | Jul 2026 | Official leaderboard |
Rankings by part
We read 4 parts from the benchmark’s paper, each ranking models on one part of the test, such as OfficeQA, BrowseComp-Plus and Terminal-Bench. Claude Opus 5 leads 3 of them, and GPT-5.6 Sol leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
OfficeQA
5 models we track have a score here.
- 1 Claude Opus 5 0.63
- 2 Kimi K3 0.59
- 3 Claude Sonnet 5 ↑1 0.53
- 4 GPT-5.6 Sol ↓1 0.49
- 5 GPT-5.6 Terra 0.14
BrowseComp-Plus
5 models we track have a score here.
- 1 Claude Opus 5 0.48
- 2 Kimi K3 0.23
- 3 Claude Sonnet 5 ↑1 0.15
- 4 GPT-5.6 Sol ↓1 0.09
- 5 GPT-5.6 Terra 0.02
Terminal-Bench
5 models we track have a score here.
- 1 Claude Opus 5 0.29
- 2 Kimi K3 0.16
- 3 Claude Sonnet 5 ↑1 0.15
- 4 GPT-5.6 Sol ↓1 0.13
- 5 GPT-5.6 Terra 0.04
GAIA
5 models we track have a score here.
- 1 GPT-5.6 Sol ↑2 0.49
- 2 Claude Opus 5 ↓1 0.47
- 3 Claude Sonnet 5 ↑1 0.33
- 4 Kimi K3 ↓2 0.31
- 5 GPT-5.6 Terra 0.30
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The HarnessOpt-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official HarnessOpt-Bench 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 AI research
- PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics.
Qwen 3.8 Max 93%
- MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
- PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
- MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
- PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
- KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup.
MiniMax-M3 28.8%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (5 entries).
- Pass Models scored 5 models we track.
- Pass Newest score The newest is from July 2026, for Claude Opus 5.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on labs.scale.com: 10 rows for models we track at the last read.
- Pass Last read November 20th, 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.