ORCA-bench
Agents investigate simulated production incidents and identify their root causes. The benchmark contains 1,079 tasks, including 324 held-out tasks covering 31 incidents.
The 2026 paper reports Claude Sonnet 4.6 as the best model for identifying all root causes, at 30.9% accuracy.
- Made by
- ORCA-bench team
- Released
- July 2026
- Kind
- Terminal tasks
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- English
- In and out
- Text, Code, Time series → Actions, Text
- Interaction
- Agentic
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- A host
- Occupations
- Computer and mathematical
- In views
- Agents
- Strongest model tested
- GPT-5.5 #21 on the capability index
- Last model added
- April 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 4.7 leads with 29.6% among models released by April 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. RCA Accuracy (Medium) on incident tasks, the official leaderboard ranking metric. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in April 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 16 more
- 1 Claude Opus 4.7 Private, reasoning effort: medium 29.6%
- 2 Claude Sonnet 4.6 Public, reasoning effort: medium 26.6%
- 3 GPT-5.5 Private, reasoning effort: medium 26.5%
- 4 GLM-5 Private 17.4%
- 5
DeepSeek-V4-Pro Public 15.6%
Source: Official ORCA-bench leaderboard
| # | Model | Score | Setting | Agent by | Released | Run on | Link | Source |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 29.6% | Private, reasoning effort: medium, Terminus 2 | Harbor | Apr 2026 | May 2026 | raw.githubusercontent.com | Official leaderboard |
| 2 | Claude Sonnet 4.6 | 26.6% | Public, reasoning effort: medium, Terminus 2 | Harbor | Feb 2026 | May 2026 | raw.githubusercontent.com | Official leaderboard |
| 3 | GPT-5.5 | 26.5% | Private, reasoning effort: medium, Terminus 2 | Harbor | Apr 2026 | May 2026 | raw.githubusercontent.com | Official leaderboard |
| 4 | GLM-5 | 17.4% | Private, Terminus 2 | Harbor | Feb 2026 | May 2026 | raw.githubusercontent.com | Official leaderboard |
| 5 | | 15.6% | Public, Terminus 2 | Harbor | Apr 2026 | May 2026 | raw.githubusercontent.com | Official leaderboard |
Rankings by part
We read 2 parts from the benchmark’s own leaderboard: RCA Accuracy (Medium) and RCA Accuracy (Hard). Each ranks models on one part of the test. Claude Opus 4.7 leads both. Arrows show how far a model moved from its overall rank among the models we track.
RCA Accuracy (Medium)
5 models we track have a score here.
- 1 Claude Opus 4.7 29.6%
- 2 Claude Sonnet 4.6 26.6%
- 3 GPT-5.5 26.5%
- 4 GLM-5 17.4%
- 5
DeepSeek-V4-Pro 15.6%
RCA Accuracy (Hard)
5 models we track have a score here.
- 1 Claude Opus 4.7 13.9%
- 2 GPT-5.5 ↑1 13.9%
- 3 Claude Sonnet 4.6 ↓1 8.96%
- 4
DeepSeek-V4-Pro ↑1 8.86%
- 5 GLM-5 ↓1 1.00%
More effort adds a median 4 points
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official ORCA-bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 4.7 | Private, reasoning effort: medium | Terminus 2 | 29.6% | ||
| Claude Sonnet 4.6 | Public, reasoning effort: medium | Terminus 2 | 26.6% | ||
| GPT-5.5 | Private, reasoning effort: medium | Terminus 2 | 26.5% | ||
| GPT-5.5 | Public, reasoning effort: medium | Terminus 2 | 24.3% | ||
| Claude Opus 4.7 | Public, reasoning effort: medium | Terminus 2 | 22.5% | ||
| Claude Sonnet 4.6 | Private, reasoning effort: medium | Terminus 2 | 22.4% | ||
| GLM-5 | Private | Terminus 2 | 17.4% | ||
| GLM-5 | Public | Terminus 2 | 16.5% | ||
| Public | Terminus 2 | 15.6% | |||
| Private | Terminus 2 | 13.3% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ORCA-bench 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 ORCA-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 Terminal agents
- SkillsBench In use 86-87 containerized expert tasks in 11 domains, run with and without Agent Skills. Verifier tests give pass rate and skill lift. GPT-5.5 66.5%
- Terminal-Bench Hard In use Agents complete 44 hard tasks from Terminal-Bench Core in the Terminus 2 harness. Artificial Analysis scores pass@1 averaged over 3 repeats.
Gemma 4 31B IT 36%
- Terminal-Bench 4.1 In use The Terminal-Bench 4.0 tasks with hardened verifiers: agents complete 66 hard tasks in containerized terminals across software, ML, science, security, and operations. Test scripts verify each result, scored as resolution rate. Copilot Fusion 71.2%
- Harbor-Index In use Agents solve about 80 hard tasks drawn from 29 benchmarks run through Harbor adapters, such as HLE, GSO and SWE-bench Pro. Each task keeps its source verifier. GPT-5.5 28.1%
- Terminal-Bench 2.1 Extended Not in model cards Agents complete long command-line tasks in containers, scored by verification tests. Mercor adds 99 new expert-written tasks to the 89 public ones and reports both scores.
Muse Wonder 54%
- Terminal-Bench Pro Not in model cards Agents solve 400 terminal tasks across eight domains. Harbor distributes the 200 public tasks. Alibaba keeps the other 200 for private evaluation. Claude Opus 4.5 51%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (5 entries).
- Pass Models scored 5 models we track.
- Warning Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
- Pass Headroom The best is 29.6%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on github.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.