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.

See the free Radar Brief
BenchLM

HLE-Verified

A verified and revised Humanity's Last Exam set that removes uncertain items and repairs fixable questions before evaluating frontier expert reasoning.

Data verified 27 confirmed releases in the last 30 daysSee the free Radar Brief

Benchmark score on HLE-Verified — September 2, 2026

We mirror the published score view for HLE-Verified. Gemini 3.8 Flash leads the public snapshot at 54.9%, followed by GPT-5.6 Sol (54.5%) and Claude Opus 5 (54.4%). We do not use these results to rank models overall.

1Proprietary

Gemini 3.8 Flash

Google

54.9%
Overall 75.37Context 1M
2Proprietary

GPT-5.6 Sol

OpenAI

54.5%
Overall 81.84Context 1.05M
3Proprietary

Claude Opus 5

Anthropic

54.4%
Overall 82.43

6 modelsKnowledgeCurrentDisplay onlyUpdated September 2, 2026

Benchmark score table (6 models)

Score
1
Gemini 3.8 FlashGoogle · Closed
54.9%
2
GPT-5.6 SolOpenAI · Closed
54.5%
3
Claude Opus 5Anthropic · Closed
54.4%
4
Gemini 3.7 FlashGoogle · Closed
53.6%
5
GPT-5.6 TerraOpenAI · Closed
51.1%
6
Claude Sonnet 5Anthropic · Closed
31.0%

The published HLE-Verified snapshot places Gemini 3.8 Flash first at 54.9%. The third row is 0.5 points behind. The broader top-10 range is 23.9 points, so the table still separates the published systems.

6 models have been evaluated on HLE-Verified. The benchmark falls in the Knowledge category. This category carries a 12% weight in BenchLM.ai's overall scoring system. HLE-Verified is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About HLE-Verified

Year

2026

Tasks

1,811 verified or revised expert questions

Format

Full-set accuracy

Difficulty

Frontier multidisciplinary expert reasoning

HLE-Verified contains 1,811 certified items and separates 689 uncertain original HLE items. Google's Gemini 3.8 evaluation reports full-set accuracy as display-only scientific-reasoning evidence; it does not overwrite the existing HLE lanes.

HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam

BenchLM freshness & provenance

Version

HLE-Verified 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 HLE-Verified measure?

A verified and revised Humanity's Last Exam set that removes uncertain items and repairs fixable questions before evaluating frontier expert reasoning.

Which model scores highest on HLE-Verified?

Gemini 3.8 Flash by Google currently leads with a score of 54.9% on HLE-Verified.

How many models are evaluated on HLE-Verified?

6 AI models have been evaluated on HLE-Verified on BenchLM.

Compare Top Models on HLE-Verified

Gemini 3.8 Flash vs GPT-5.6 SolGPT-5.6 Sol vs Claude Opus 5Claude Opus 5 vs Gemini 3.7 FlashGemini 3.7 Flash vs GPT-5.6 Terra
Last updated: September 2, 2026 · BenchLM version HLE-Verified 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.