Minimal-LinuxBench (Anthropic)
Agents attempt a hidden side task in high-stakes software environments while a monitor watches. Score is the stealth success rate.
Anthropic adaptation of a subset of LinuxBench, an AI control arena that Redwood Research shared privately.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Claude Opus 4.7 leads with 2.60% among models released by May 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in May 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 15 more
- 1 Claude Opus 4.7 2.60%
- 2 Claude Mythos Preview 2.70%
- 3 Claude Opus 4.8 4.70%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 2.60% | Apr 2026 | Model card | |
| 2 | Claude Mythos Preview | 2.70% | Apr 2026 | Model card | |
| 3 | Claude Opus 4.8 | 4.70% | May 2026 | Model card |
2 models have held number one since its release in Apr 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
3 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 28 May 2026 | Claude Opus 4.8 system card | 4.7 | Minimal-LinuxBench stealth success rate, Opus 4.6 monitor without the chain of thought, about 170 tasks (2.7% when the monitor sees the chain of thought, the deployed setting), printed on chart (6.4.2.2, Figure 6.4.2.2.A) | Monitorability |
| 16 Apr 2026 | Claude Opus 4.7 system card | 2.6 | Minimal-LinuxBench stealth success rate against the Opus 4.6 monitor using CoT (the current monitor), printed on chart. 3.1 against the no-CoT monitor (6.4.2.2, Figure 6.4.2.2.A) | Monitorability |
| 7 Apr 2026 | Claude Mythos Preview system card | 2.7 | Minimal-LinuxBench stealth success rate, no extended thinking, current Opus 4.6 monitor with CoT, 200 sampled task pairs, value printed on chart; 11.4 with the no-CoT monitor, side task completed in 69 of 200 (4.4.3.2, Figure 4.4.3.2.A) | Monitorability |
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% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy.
Muse Spark Open 98.1% - 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. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%