Best AI for Fine-Tuning
Compare open-weights AI models for fine-tuning — models you can train on your data and deploy on your own infrastructure.Rankings combine multiple benchmark scores weighted by relevance to this use case. Updated August 2026.
Why This Matters
Fine-Tuning is one of the most common AI workloads — and the best model depends heavily on the specific task. A model that excels at creative writing may struggle with structured data extraction, and vice versa.
We built this use-case ranking by combining multiple benchmark categories with weights tuned to match real-world usage patterns. For fine-tuning, the score emphasizes the benchmarks that matter most: task accuracy, output quality, and consistency. Speed and cost are factored in but secondary to quality.
All scores are from public independent benchmarks. For a broader view across all tasks, see the overall LLM leaderboard or the expert picks page.
Quick Answer
The best AI models for fine-tuning are:1. GLM-5.1 (Z.ai, score: 72.0), 2. Qwen3.8 Flash Next (Alibaba, score: 61.5), 3. Kimi K3 (Moonshot AI, score: 59.7).
| # | Model | Provider | Score | Speed | Price (output) |
|---|---|---|---|---|---|
| 1 | GLM-5.1 | Z.ai | 72.0 | 74 t/s | $4.4/M |
| 2 | Qwen3.8 Flash Next | Alibaba | 61.5 | 168 t/s | $0.4/M |
| 3 | Kimi K3 | Moonshot AI | 59.7 | 37 t/s | $15/M |
| 4 | GLM-5.3 Flash | Z.ai | 58.0 | 182 t/s | $0.3/M |
| 5 | DeepSeek V4 Pro | DeepSeek | 53.2 | 66 t/s | $0.87/M |
| 6 | GLM-5.2 | Z.ai | 52.6 | 164 t/s | $4.29/M |
| 7 | DeepSeek V4 Flash | DeepSeek | 51.8 | 113 t/s | $0.28/M |
| 8 | Granite 4.2 30B Instruct | IBM | 51.0 | 78 t/s | $1.2/M |
| 9 | GPT-oss-120B | OpenAI | 47.0 | 240 t/s | $0.4/M |
| 10 | Kimi K2.6 | Moonshot AI | 45.1 | 40 t/s | $4/M |
| 11 | Qwen3 8 2 4t A95b | Alibaba | 45.0 | 85 t/s | $2/M |
| 12 | Glm 5 | Z.ai | 45.0 | 60 t/s | $3/M |
| 13 | Deepseek V4 Pro 0424 | DeepSeek | 45.0 | 60 t/s | $1.1/M |
| 14 | Deepseek V4 Pro 0424 High | DeepSeek | 45.0 | 60 t/s | $1.1/M |
| 15 | Qwen3 6 Max | Alibaba | 45.0 | 85 t/s | $2/M |