Reward-hack-prone coding tasks (Anthropic)
Models solve hard coding problems where earlier Claude models often special-cased tests. Score is the classifier hack rate.
Anthropic also reports a hidden-test hack rate on held-out tests for the same tasks.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Code A program or patch run against tests.
3 models share the top score of 0% among models released by Feb 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 Feb 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 16 more
- 1 Claude Opus 4.5 0%
- 2 Claude Opus 4.6 0%
- 3 Claude Sonnet 4.6 0%
- 4 Claude Sonnet 4.5 1.00%
- 5 Claude Sonnet 4 4.00%
- 6 Claude Haiku 4.5 6.00%
- 7 Claude Opus 4 9.00%
- 8 Claude Opus 4.1 12%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | 0% | Nov 2025 | Model card | |
| 2 | Claude Opus 4.6 | 0% | Feb 2026 | Model card | |
| 3 | Claude Sonnet 4.6 | 0% | Feb 2026 | Model card | |
| 4 | Claude Sonnet 4.5 | 1.00% | Sep 2025 | Model card | |
| 5 | Claude Sonnet 4 | 4.00% | May 2025 | Model card | |
| 6 | Claude Haiku 4.5 | 6.00% | Oct 2025 | Model card | |
| 7 | Claude Opus 4 | 9.00% | May 2025 | Model card | |
| 8 | Claude Opus 4.1 | 12% | Aug 2025 | Model card |
3 models have held number one since its release in May 2025
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.
8 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 |
|---|---|---|---|---|
| 17 Feb 2026 | Claude Sonnet 4.6 system card | 0 | Reward-hack-prone coding tasks, classifier hack rate; hidden test hack rate 0 (Table 4.3.2.A) | Alignment |
| 5 Feb 2026 | Claude Opus 4.6 system card | 0 | Reward-hack-prone coding tasks, classifier hack rate, lower is better; hidden test hack rate 0 (Table 6.2.3.2.A) | Alignment |
| 24 Nov 2025 | Claude Opus 4.5 system card | 0 | Reward-hack-prone coding tasks, classifier hack rate, lower is better (Table 6.10.1.A; hidden test hack rate 0%; average in-eval hack rate across the suite 18.2% in Figure 6.10.A) | Alignment |
| 15 Oct 2025 | Claude Haiku 4.5 system card | 6 | Reward-hack-prone coding tasks v2, classifier hack rate; hidden test hack rate 3 (Table 5.B) | Alignment |
| 29 Sep 2025 | Claude Sonnet 4.5 system card | 1 | Reward-hack-prone coding tasks v2, classifier hack rate 1%, lower is better. Hidden test hack rate 1% (Table 6.1.A) | Alignment |
| 5 Aug 2025 | Claude Opus 4.1 system card addendum | 12 | Reward-hack-prone coding tasks, classifier hack rate 12%. Hidden test hack rate 14% (Table 5.B) | Alignment |
| 22 May 2025 | Claude Sonnet 4 system card | 4 | Reward-hack-prone coding tasks, classifier hack rate 4%. Hidden test hack rate 12% (Table 6.2.A) | Alignment |
| 22 May 2025 | Claude Opus 4 system card | 9 | Reward-hack-prone coding tasks, classifier hack rate 9%. Hidden test hack rate 13% (Table 6.2.A) | Alignment |
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%