Benchmark profile

Scientific Code Benchmark (SciCode)

SciCode evaluates language models on generating code for realistic scientific research problems across 16 subfields of physics, math, chemistry, biology, and material science. Problems decompose into 338 subproblems requiring domain knowledge recall, scientific reasoning, and precise code synthesis. Based on real scripts from published research.

Data verified July 23, 2026

Top models on SciCode — July 23, 2026

As of July 23, 2026, Sakana Fugu leads the SciCode leaderboard with 60.1% , followed by Sakana Fugu-Ultra (58.7%) and Qwen3.7 Max (53.5%).

1Closed

Sakana Fugu

Sakana AI

sakana-fugu

60.1%

Overall —Context 1M

2Closed

Sakana Fugu-Ultra

Sakana AI

sakana-fugu-ultra

58.7%

Overall —Context 1M

3Closed

Qwen3.7 Max

Alibaba

qwen3-7-max

53.5%

Overall 72.84Context 1M

13 modelsCoding16% of category scoreRefreshingUpdated July 23, 2026

Leaderboard (13 models)

Score

1

Sakana FuguSakana AI · Closed

60.1%

2

Sakana Fugu-UltraSakana AI · Closed

58.7%

3

Qwen3.7 MaxAlibaba · Closed

53.5%

4

Gemini 3.5 FlashGoogle · Closed

53.1%

5

Kimi K2.6Moonshot AI · Open weight

52.2%

6

Qwen3.7 PlusAlibaba · Closed

51.3%

7

Kimi K2.5Moonshot AI · Open weight

48.7%

8

Grok 4.3xAI · Closed

47.3%

9

Qwen 3.6 Max (preview)Alibaba · Closed

47%

10

Nemotron 3 UltraNVIDIA · Open weight

44.6%

11

Hy3 PreviewTencent · Open weight

41.2%

12

Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight

32%

13

Ling 2.6 FlashInclusionAI · Open weight

27%

According to BenchLM.ai, Sakana Fugu leads the SciCode benchmark with a score of 60.1%, followed by Sakana Fugu-Ultra (58.7%) and Qwen3.7 Max (53.5%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.

13 models have been evaluated on SciCode. The benchmark falls in the Coding category. This category carries a 20% weight in BenchLM.ai's overall scoring system. Within that category, SciCode contributes 16% of the category score, so strong performance here directly affects a model's overall ranking.

About SciCode

Year

2024

Tasks

80

BenchLM freshness & provenance

Version

SciCode 2024

Refresh cadence

Annual

Staleness state

Refreshing

Question availability

Public benchmark set

Refreshing

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 SciCode measure?

SciCode evaluates language models on generating code for realistic scientific research problems across 16 subfields of physics, math, chemistry, biology, and material science. Problems decompose into 338 subproblems requiring domain knowledge recall, scientific reasoning, and precise code synthesis. Based on real scripts from published research.

Which model scores highest on SciCode?

Sakana Fugu by Sakana AI currently leads with a score of 60.1% on SciCode.

How many models are evaluated on SciCode?

13 AI models have been evaluated on SciCode on BenchLM.

Compare Top Models on SciCode

Sakana Fugu vs Sakana Fugu-Ultra Sakana Fugu-Ultra vs Qwen3.7 Max Qwen3.7 Max vs Gemini 3.5 Flash Gemini 3.5 Flash vs Kimi K2.6

Last updated: July 23, 2026 · BenchLM version SciCode 2024

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.