Tool calling models, ranked
An agent lives or dies on two things: calling the right function with valid arguments, and returning JSON your code can parse without a retry. Everything here does both. In a multi-step loop, a cheaper model with a low error rate beats an expensive one with the same error rate, because every retry pays twice.
Most used in this category
- 1GPT-4o-mini118 tok174,075
Best value in this category
Score out of 100: 50% cheapness, 30% capabilities, 20% context size — not a quality benchmark.
| # | Model | Context | ≈ 1,000 words | Score |
|---|---|---|---|---|
| 1 | Gemini 2.5 Flash Lite (batch) | 1,048,576 | 94 | 85.8 |
| 2 | Qwen3.7 Flash | 1,000,000 | 60 | 84.5 |
| 3 | Muse Spark 1.2 Contributor | 1,048,576 | 113 | 83.8 |
| 4 | Muse Spark 1.3 Contributor | 1,048,576 | 113 | 83.8 |
| 5 | GPT-5 Nano (batch) | 400,000 | 85 | 83.5 |
| 6 | Nex-N2-Mini | 262,144 | 47 | 82.5 |
| 7 | Ling 3.0 Flash | 262,144 | 32 | 80.7 |
| 8 | GPT-4.1 Nano (batch) | 1,047,576 | 94 | 79.8 |
| 9 | GLM Flash Latest | 1,310,720 | 116 | 78.3 |
| 10 | Gemini 2.5 Flash Lite | 1,048,576 | 189 | 78.2 |
| 11 | GLM 5.3 Flash | 1,310,720 | 123 | 77.6 |
| 12 | Solar Pro 4 | 524,288 | 57 | 76.8 |
| 13 | Qwen3.5-Flash | 1,000,000 | 123 | 76.7 |
| 14 | DeepSeek V4 Flash Latest | 1,310,720 | 79 | 76.4 |
| 15 | GPT-5 Nano | 400,000 | 170 | 76 |
Full list of models that call your tools reliably with prices →