Coding benchmark
LiveCodeBench v6 leaderboard
Every model the catalog carries a published LiveCodeBench v6 value for, ranked by that value.
LiveCodeBench v6 is a named release slice used in provider comparison tables. Keeping it separate prevents v6 results from being mixed into older or rolling LiveCodeBench windows.
LiveCodeBench v6 ranking
31 models with a published LiveCodeBench v6 value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Provider-published v6 competitive programming results |
|---|---|---|---|
| 1 | SASakana Fugu-Ultra | Sakana AI | 93.2 |
| 2 | SASakana Fugu | Sakana AI | 92.9 |
| 3 |  Qwen3.8-Omni-Flash | Alibaba | 92.6 |
| 4 |  Solar Open 2 | Upstage | 92.4 |
| 5 | Qwen3.8-Flash-Next | Alibaba | 91.9 |
| 6 | DSdots3-note Preview | Dots Studio | 91.5 |
| 7 | Qwen3.8-27B | Alibaba | 90.3 |
| 8 | PMTernary Bonsai 2 27B | Prism ML | 90.1 |
| 9 |  Kimi K2.6 | Moonshot AI | 89.6 |
| 10 |  Nemotron 3 Ultra | NVIDIA | 89.0 |
| 11 | BTBTL-3 | Bad Theory Labs | 88.1 |
| 12 |  MAI-Thinking-1 | Microsoft | 87.7 |
| 13 | Qwen3.6 Plus | Alibaba | 87.1 |
| 14 | Kimi K2.5 | Moonshot AI | 85.0 |
| 15 |  Claude Opus 4.5 | Anthropic | 84.8 |
| 16 | STA.X K2 | SK Telecom | 84.0 |
| 17 | Qwen3.5 397B | Alibaba | 83.6 |
| 18 |  Granite 4.2 30B | IBM | 75.8 |
| 19 | Granite 4.2 8B | IBM | 73.2 |
| 20 |  Gemma 4 12B | 72.0 | |
| 21 | JEMellum2-12B-A2.5B-Thinking | JetBrains | 69.9 |
| 22 | Granite 4.2 3B | IBM | 69.7 |
| 23 | OPMiniCPM5-2B | OpenBMB | 69.1 |
| 24 | BTBTL-4 | Bad Theory Labs | 66.1 |
| 25 | ZYZAYA1-8B | Zyphra | 65.8 |
| 26 | ZYZAYA1-74B-Preview | Zyphra | 65.7 |
| 27 |  Agents-A1-4B | InternScience | 59.6 |
| 28 |  LFM2.5-2.6B | LiquidAI | 59.4 |
| 29 | JEMellum2-12B-A2.5B-Instruct | JetBrains | 37.2 |
| 30 | OPMiniCPM5-1B | OpenBMB | 33.5 |
| 31 |  LLaDA2.2-mini | InclusionAI | 28.1 |
Evidence key: Observed
Rows are ordered by the value LiveCodeBench official repository and release documentation published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.