DeepSeek-V2.5: benchmark scores, price and ranking
DeepSeek-V2.5 from deepseek-ai ranks #162 of 743 LLMs on the LLMs Tiger aggregate benchmark score (50.9), based on 2 benchmark results captured up to 2026-10-13.
- Organisation
- deepseek-ai
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2024-09-05 (source: curated)
- Aggregate rank
- #162 of 743 (score 50.9)
Across the 2 benchmarks where at least ten models are scored, DeepSeek-V2.5 is in the top 10% on 1 and in the top half on 2.
DeepSeek-V2.5 benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| Aider Code Editing | 72.2% | #9 of 92 | Aider Aider Code Editing leaderboard | 2026-10-13 |
| MMLU-Pro | 65.8% | #49 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
About DeepSeek-V2.5
DeepSeek-V2.5 is an open-weight model that can run on your own hardware from deepseek-ai. It was released on 2024-09-05, according to the developer's announcement. No API price is listed for it yet. We hold 2 benchmark results for it across 2 areas, and 1 is in the top tenth of its board.
Where DeepSeek-V2.5 stands
On our aggregate score, DeepSeek-V2.5 is #162 of 743 ranked models, in the top quarter. Its score is 50.9; the scale is centred on 50, so it sits 0.9 points above the typical model. That beats 78% of the ranked field. Among the 7 ranked deepseek-ai models it is #2. The next one up is DeepSeek R1 (0528) at 53.6. The next one down is DeepSeek-V2-Chat at 47.8.
DeepSeek-V2.5, benchmark by benchmark
On Aider Code Editing, DeepSeek-V2.5 scores 72.2% and ranks #9 of 92. Only 8 models score higher, led by Claude 3.5-Sonnet-20241022 at 84.2%, which is 12.0 points ahead. It covers Aider official benchmark. The board median is 55.6%, so DeepSeek-V2.5 is 16.6 points above it. Source: Aider Aider Code Editing leaderboard, captured 2026-10-13.
On MMLU-Pro, DeepSeek-V2.5 reaches 65.8%. It ranks #49 of 136, ahead of 64% of the field, and 22.3 points behind Darwin-180B-RSI at 88.1%. The results either side are Mistral-Large-Instruct-2407 (65.9%) and NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16 (65.1%). It covers 57-subject knowledge + reasoning. The board median is 55.3%, so DeepSeek-V2.5 is 10.6 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
Strengths and weaknesses
DeepSeek-V2.5 has too few results on boards with ten or more models to say where it is strong or weak. Treat any single score as a data point, not a verdict.
Release history
The deepseek-ai model released before it was DeepSeek-Coder-V2-Lite-Base on 2024-06-13, which scores 45.0.
Reading these numbers
The 2 scores on this page come from 2 sources: Aider Aider Code Editing leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). 1 of them come from an official leaderboard and 1 from datasets or third-party runs. With 2 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #159 LongCat-Flash-Thinking-2601 (50.9)
- #160 MiniMax-M1-80k (50.9)
- #161 Optimus Alpha (50.9)
- #163 Qwen3.8 Max (0902) (xhigh) (50.8)
- #164 GPT-4 (0613) (50.8)
- #165 Claude Code + GLM-5.3 (50.8)
More from deepseek-ai
- DeepSeek R1 (0528) (#33, 53.6)
- DeepSeek-V2-Chat (#404, 47.8)
- DeepSeek-V2 (#505, 46.6)
- DeepSeek-Coder-V2-Lite-Instruct (#546, 46.0)
- deepseek-math-7b-instruct (#604, 45.2)
- DeepSeek-Coder-V2-Lite-Base (#608, 45.0)
Compare it on the live leaderboard → · How the aggregate score works