Codesota · Reasoning · Multi-step Reasoning · HLETasks/Reasoning/Multi-step Reasoning
Multi-step Reasoning · benchmark dataset · 2025 · EN

Humanity's Last Exam.

3,000 expert-level questions designed to be the hardest public benchmark. Questions sourced from domain experts across mathematics, sciences, humanities, and more. Frontier difficulty — most models score below 10%.

Paper ↗Submit a result ↵
§ 01 · Leaderboard

Best published scores.

13 results indexed across 1 metric. Shaded row marks current SOTA; ties broken by submission date.

Primary
accuracy · higher is better
accuracy· primary
13 rows
#ModelOrgSubmittedPaper / codeaccuracy
01Gemini 3 ProAPIGoogle——38.30
02GPT-5APIOpenAI——25.30
03Grok 4APIxAI——24.50
04Gemini 2.5 ProGoogle——21.60
05GPT-5-miniOpenAI——19.40
06Claude Opus 4.6APIAnthropicApr 2026scale-ai-leaderboard19
07Claude 4.5 SonnetAnthropic——13.70
08Claude Sonnet 4.6APIAnthropicApr 2026pricepertoken-leaderboard13.20
09Gemini 2.5 FlashGoogle——12.10
10DeepSeek R1OSSDeepSeek——8.50
11o1APIOpenAI——8.00
12GPT-4.1 miniAPIOpenAIApr 2026pricepertoken-leaderboard4.60
13GPT-4oAPIOpenAI——2.70
Fig 2 · Rows sorted by score within each metric. Shaded row marks SOTA. Dates reflect model or paper release where available, otherwise the date Codesota accessed the source.
§ 03 · Progress

1 steps
of state of the art.

Each row below marks a model that broke the previous record on accuracy. Intermediate submissions are kept in the leaderboard above; only SOTA-setting entries are re-listed here.

Higher scores win. Each subsequent entry improved upon the previous best.

SOTA line · accuracy
  1. Apr 12, 2026Claude Opus 4.6Anthropic19
Fig 3 · SOTA-setting models only. 1 entries span Apr 2026 → Apr 2026.
§ 06 · Contribute

Have a score that beats
this table?

Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.

Submit a result ↵Read submission guide
What a submission needs
  • 01A public checkpoint or API endpoint
  • 02A reproduction script with frozen commit + seed
  • 03Declared evaluation environment (Python, deps)
  • 04One row per metric declared by this dataset
  • 05A contact so we can follow up on discrepancies