LiveCodeBench Pro
We show this table for reference; we do not rank on it.
A harder competitive-programming benchmark family built from Codeforces, ICPC, and IOI problems, with quarter-specific public leaderboards and difficulty-aware reporting.
Benchmark score on LiveCodeBench Pro — November 13, 2026
We compile the LiveCodeBench Pro rows from provider self-reports and secondary reports. Sakana Fugu-Ultra leads the table at 90.8%, followed by Sakana Fugu (87.8%) and GPT-5.4 (87.5%). We do not use these results to rank models overall.
Sakana Fugu-Ultra
Sakana AI
Reported by Sakana AISakana Fugu
Sakana AI
Reported by Sakana AIGPT-5.4
OpenAI
Reported by Meta AI (comparison table)9 modelsCodingCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (9 models)
None of the 9 results on this page has a documented run by the benchmark or an independent evaluator: 5 reported by the labs that built the models and 4 reported in another publisher’s document.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Sakana Fugu-UltraSakana AIReported by Sakana AI | 90.8% | Reported by Sakana AI | Not reported | Closed |
| 2 | Sakana FuguSakana AIReported by Sakana AI | 87.8% | Reported by Sakana AI | Not reported | Closed |
| 3 | GPT-5.4OpenAIReported by Meta AI (comparison table) | 87.5% | Reported by Meta AI (comparison table) | Not reported | Closed |
| 4 | Gemini 3.1 ProGoogleReported by Meta AI (comparison table) | 82.9% | Reported by Meta AI (comparison table) | Not reported | Closed |
| 5 | Muse SparkMetaReported by Meta | 80.0% | Reported by Meta | Not reported | Closed |
| 6 | Grok 4.20xAIReported by Meta AI (comparison table) | 74.2% | Reported by Meta AI (comparison table) | Not reported | Closed |
| 7 | Claude Opus 4.6AnthropicReported by Meta AI (comparison table) | 70.7% | Reported by Meta AI (comparison table) | Not reported | Closed |
| 8 | LFM3-1.7BLiquidAIReported by LiquidAI | 31.7% | Reported by LiquidAI | Not reported | Open |
| 9 | MiniCPM5-1BOpenBMBReported by OpenBMB | 22.7% | Reported by OpenBMB | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Among the reported LiveCodeBench Pro rows, Sakana Fugu-Ultra is first at 90.8%. The third row is 3.3 points behind. The broader top-10 range is 68.1 points, so the table still separates the published systems.
9 models have been evaluated on LiveCodeBench Pro. The benchmark falls in the Coding category. LiveCodeBench Pro is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About LiveCodeBench Pro
Year
2025
Tasks
Quarter-specific contest programming sets
Format
Competitive programming
Difficulty
High-end contest programming
LiveCodeBench Pro is distinct from the original LiveCodeBench family. It excludes LeetCode, emphasizes stronger contest difficulty, and the official site publishes quarter-specific leaderboards such as 25Q2 with hard, medium, and easy pass rates.
Freshness and provenance
Version
LiveCodeBench Pro 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
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.
Questions
What does LiveCodeBench Pro measure?
A harder competitive-programming benchmark family built from Codeforces, ICPC, and IOI problems, with quarter-specific public leaderboards and difficulty-aware reporting.
Which model scores highest on LiveCodeBench Pro?
Sakana Fugu-Ultra by Sakana AI currently leads with a score of 90.8% on LiveCodeBench Pro.
How many models are evaluated on LiveCodeBench Pro?
9 AI models have published results on LiveCodeBench Pro in the BenchLM catalog.
Compare top models on LiveCodeBench Pro
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.