BARC-Bench
Agents handle routine workplace requests and use records to infer privacy and safety constraints. A judge checks whether each attempt completes safely or defers.
The authors tested four model configurations five times on each of 110 tasks in nine simulated workplaces.
- Made by
- Scale AI
- Released
- October 2026
- Kind
- Deception and alignment
- Level
- University
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Measures
- Propensity
- Risk
- Privacy
- Behavior
- Lawlessness, Misaligned with intent
- Language
- English
- Interaction
- Agentic
- Environment
- A simulator
- Items from
- Written by experts
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents, Robotics, Content safety
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
GPT-6 Astra leads with 72.7% among models released by September 2026
Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in September 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
- Claude Opus 5.5
- and 14 more
Source: The BARC-Bench paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 72.7% | Default thinking budget, OpenCode | Sep 2026 | Official leaderboard |
| 2 | Claude Fable 5.1 | 37.3% | Default thinking budget, OpenCode | Sep 2026 | Official leaderboard |
| 3 | | 23.6% | Default thinking budget, OpenCode | Sep 2026 | Official leaderboard |
| 4 | | 9.10% | Default thinking budget, OpenCode | Apr 2026 | Official leaderboard |
Rankings by part
We read 8 parts from the benchmark’s paper, each ranking models on one part of the test, such as Conflict: Communication, Conflict: File operations and Conflict: Workspace applications. GPT-6 Astra leads 7 of them, and Claude Fable 5.1 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
Conflict: Communication
4 models we track have a score here.
- 1 Claude Fable 5.1 ↑1 44.4%
- 2 GPT-6 Astra ↓1 44.4%
- 3
Gemini 3.8 Flash 22.2%
- 4
DeepSeek-V4-Pro 0%
Conflict: File operations
4 models we track have a score here.
- 1 GPT-6 Astra 83.3%
- 2 Claude Fable 5.1 33.3%
- 3
DeepSeek-V4-Pro ↑1 16.7%
- 4
Gemini 3.8 Flash ↓1 0%
Conflict: Workspace applications
4 models we track have a score here.
- 1 GPT-6 Astra 50%
- 2 Claude Fable 5.1 25%
- 3
DeepSeek-V4-Pro ↑1 0%
- 4
Gemini 3.8 Flash ↓1 0%
Conflict: All
4 models we track have a score here.
- 1 GPT-6 Astra 57.9%
- 2 Claude Fable 5.1 36.8%
- 3
Gemini 3.8 Flash 10.5%
- 4
DeepSeek-V4-Pro 5.30%
No conflict: Communication
4 models we track have a score here.
- 1 GPT-6 Astra 65.6%
- 2 Claude Fable 5.1 26.2%
- 3
Gemini 3.8 Flash 19.7%
- 4
DeepSeek-V4-Pro 4.90%
No conflict: File operations
4 models we track have a score here.
- 1 GPT-6 Astra 96.4%
- 2 Claude Fable 5.1 60.7%
- 3
Gemini 3.8 Flash 39.3%
- 4
DeepSeek-V4-Pro 21.4%
No conflict: Workspace applications
4 models we track have a score here.
- 1 GPT-6 Astra 100%
- 2 Claude Fable 5.1 50%
- 3
Gemini 3.8 Flash 50%
- 4
DeepSeek-V4-Pro 0%
No conflict: All
4 models we track have a score here.
- 1 GPT-6 Astra 75.8%
- 2 Claude Fable 5.1 37.4%
- 3
Gemini 3.8 Flash 26.4%
- 4
DeepSeek-V4-Pro 9.90%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The BARC-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 BARC-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 Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6%
- Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
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 (4 entries).
- Note Models scored Only 4 models we track.
- Pass Newest score The newest is from September 2026, for GPT-6 Astra.
- Note Frontier models Only 2 of the top 10 have a score. The strongest tested is GPT-6 Astra (#4).
- Pass Headroom The best is 72.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 4 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.