SWE-bench Multilingual Leaderboard
coding13 models · Updated October 2026
On the SWE-bench Multilingual benchmark, Gemini 3 Flash ranks #1 with a score of 72.7, while DeepSeek V3.2 offers the best score-per-dollar at $0.80/1M output tokens. The full ranking, with cost per million tokens, is below.
💰 Best value
DeepSeek V3.2 — score 59.0 at $0.80/1M output tokens
| # | Model | Score | Output / 1M | Context |
|---|---|---|---|---|
| 1 | Gemini 3 FlashOpenCode | 72.7 | $3.00 | 1M |
| 2 | Claude Opus 4.6Anthropic | 72.0 | $25.00 | 1M |
| 3 | Claude Opus 4.5Anthropic | 70.7 | $25.00 | 200K |
| 4 | GLM 5Z.ai | 69.7 | $3.20 | 205K |
| 5 | Gemini 3 ProOpenCode | 68.7 | $12.00 | 1M |
| 6 | MiniMax 2.5Regolo AI | 68.3 | $3.50 | 190K |
| 7 | Kimi K2.5MoonshotAI | 67.3 | $3.00 | 262K |
| 8 | Claude Sonnet 4.5Anthropic | 67.0 | $15.00 | 1M |
| 9 | GPT-5.2OpenAI | 66.7 | $14.00 | 400K |
| 10 | GPT-5.2-CodexOpenAI | 66.3 | $14.00 | 400K |
| 11 | Claude Haiku 4.5Anthropic | 64.7 | $5.00 | 200K |
| 12 | DeepSeek V3.2DeepSeek | 59.0 | $0.80 | 164K |
| 13 | GPT-5 MiniOpenAI | 39.7 | $2.00 | 400K |
What does SWE-bench Multilingual test?
Evaluation framework with 2,294 real GitHub software-engineering problems.
Frequently asked questions
What is the SWE-bench Multilingual benchmark?
Evaluation framework with 2,294 real GitHub software-engineering problems.
Which model is best on SWE-bench Multilingual?
Gemini 3 Flash (OpenCode) currently leads the SWE-bench Multilingual leaderboard with a score of 72.7.
What is the best value model on SWE-bench Multilingual?
DeepSeek V3.2 offers the best score-per-dollar on SWE-bench Multilingual — a score of 59.0 at $0.80 per 1M output tokens.
Pricing is indicative — confirm with the provider before production use. Updated October 2026.