Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start free brief

Benchmark profile

HackerBench v0.2 benign refusal rate (HackerBench benign refusal)

The rate at which a model refuses benign cyber requests in xAI's HackerBench v0.2 evaluation.

Data verified 26 confirmed releases in the last 30 daysStart free brief

Benchmark score on HackerBench benign refusal — August 12, 2026

BenchLM mirrors the published score view for HackerBench benign refusal. GPT-5.6 Sol leads the public snapshot at 0.0% , followed by Claude Opus 4.8 (0.0%) and Grok 4.5 (0.0%). We do not use these results to rank models overall.

5 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated August 12, 2026

Benchmark score table (5 models)

Score
1
GPT-5.6 SolOpenAI · Closed
0.0%
2
Claude Opus 4.8Anthropic · Closed
0.0%
3
Grok 4.5xAI · Closed
0.0%
4
Grok 4.6xAI · Closed
0.8%
5
GPT-5.5OpenAI · Closed
1.1%

The published HackerBench benign refusal snapshot places GPT-5.6 Sol first at 0.0%. The third row is 0.0 points higher. The broader top-10 range is 1.1 points, so many of the published results sit in a relatively narrow band.

5 models have been evaluated on HackerBench benign refusal. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. HackerBench benign refusal is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About HackerBench benign refusal

Year

2026

Tasks

Benign cyber prompts

Format

Refusal rate

Difficulty

Cyber helpfulness

No public HackerBench v0.2 repository or result page was found. Lower refusal is better; the lane is display-only.

BenchLM freshness & provenance

Version

HackerBench benign refusal 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does HackerBench benign refusal measure?

The rate at which a model refuses benign cyber requests in xAI's HackerBench v0.2 evaluation.

Which model scores highest on HackerBench benign refusal?

GPT-5.6 Sol by OpenAI currently leads with a score of 0.0% on HackerBench benign refusal.

How many models are evaluated on HackerBench benign refusal?

5 AI models have been evaluated on HackerBench benign refusal on BenchLM.

Last updated: August 12, 2026 · BenchLM version HackerBench benign refusal 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.