AlgoTune
Speeding up numeric code beyond reference implementations.
As of 19 Sept 2026, GPT-5.2 leads AlgoTune on BenchLeader with 2.0, ahead of Gemini 3.1 Pro at 2.0, across 18 model configurations with a published result.
- Published by
- AlgoTunedata via Epoch AI Benchmarking Hub
- Category
- Coding
- Index weight
- Reference only
- Models
- 18
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
The model optimises functions and is scored on speed-up versus the reference.
How it is scored
Speed-up score, as published by AlgoTune.
What to keep in mind
A speed-up multiple, not a percent.
- 1GPT-5.2 (medium)2.0
- 2Gemini 3.1 Pro2.0
- 3GPT-5.4 (high)1.9
- 4Gemini 3 Pro1.8
- 5Claude Opus 4.51.8
- 6o4-mini (high)1.7
- 7DeepSeek R11.7
- 8GPT-5 (high)1.7
- 9Claude Sonnet 4.51.5
- 10GLM-4.5 (thinking)1.5
- 11Gemini 2.5 Pro1.5
- 12Claude Opus 4.6 (high)1.5
- 13Qwen3 Coder 480B1.4
- 14gpt-oss-120b (high)1.4
- 15GPT-5 mini (high)1.4
18 of 18
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-5.2mediumOpenAI | 2.0 | 58.7 | 2025-12-11 |
| 2 |  Gemini 3.1 ProGoogle | 2.0 | 63.9 | 2026-02-19 |
| 3 | GPT-5.4highOpenAI | 1.9 | 59.0 | 2026-03-05 |
| 4 | Gemini 3 ProGoogle | 1.8 | 61.1 | 2025-11-18 |
| 5 |  Claude Opus 4.5Anthropic | 1.8 | 58.2 | 2025-11-24 |
| 6 | o4-minihighOpenAI | 1.7 | 50.8 | 2025-04-16 |
| 7 |  DeepSeek R1DeepSeekopen | 1.7 | 48.7 | 2025-01-20 |
| 8 | GPT-5highOpenAI | 1.7 | 58.6 | 2025-08-07 |
| 9 | Claude Sonnet 4.5Anthropic | 1.5 | 54.3 | 2025-09-29 |
| 10 |  GLM-4.5thinkingZhipu AIopen | 1.5 | 48.3 | 2025-08-05 |
| 11 | Gemini 2.5 ProGoogle | 1.5 | 54.5 | 2025-06-17 |
| 12 | Claude Opus 4.6highAnthropic | 1.5 | 61.3 | 2026-02-05 |
| 13 |  Qwen3 Coder 480BAlibabaopen | 1.4 | 48.8 | 2025-07-31 |
| 14 | gpt-oss-120bhighOpenAIopen | 1.4 | 49.7 | 2025-08-05 |
| 15 | GPT-5 minihighOpenAI | 1.4 | 55.3 | 2025-08-07 |
| 16 | Claude Opus 4.1Anthropic | 1.3 | 51.9 | 2025-08-05 |
| 17 | Claude Opus 4Anthropic | 1.3 | 52.9 | 2025-05-22 |
| 18 | GPT-5 PromediumOpenAI | 1.3 | – | 2025-10-07 |
Cite as: BenchLeader, “AlgoTune leaderboard”, https://www.benchleader.com/benchmarks/algotune, data as of 19 Sept 2026.
AlgoTune: questions
- What does AlgoTune measure?
- The model optimises functions and is scored on speed-up versus the reference. Scores are reported in score; higher is better.
- Which AI model leads AlgoTune?
- GPT-5.2 leads AlgoTune with 2.0 as of 19 Sept 2026, ahead of Gemini 3.1 Pro at 2.0.
- How many models have AlgoTune results?
- 18 model configurations have a AlgoTune result on BenchLeader, all taken from AlgoTune via Epoch AI Benchmarking Hub.
- Who runs AlgoTune and how often is it updated?
- AlgoTune is published by AlgoTune. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does AlgoTune count toward the BenchLeader Index?
- No. AlgoTune is shown for reference but left out of the composite index.