Best LLMs for agents & tool use
Multi-step agents, tool calling, long-horizon workflows. Every model in the catalog, ranked three ways and priced at a realistic 30M input / 8M output tokens a month for this task.
What the engine picks on the default cost/quality/speed mix before you touch a slider.
Top task score (99/100) at $350/mo for this volume.
The cheapest model that stays close to the leader on quality — 90/100 at $38.0/mo.
By quality
Highest task score first.
Best value
Most quality per dollar of monthly spend.
Cheapest
Lowest monthly cost at this volume.
Fit to your volume
The ranking above prices a typical month. Enter your own agents & tool use volume and, if you like, a monthly ceiling — the picks re-rank as you type. No signup.
| Model | Quality | Per month | Value |
|---|---|---|---|
| GPT-5 nano OpenAI | 68 | $4.70 | 14.47 |
| Mistral Small 3.2 Mistral | 68 | $5.40 | 12.59 |
| Gemini 2.5 Flash-Lite Google | 62 | $6.20 | 10.00 |
| Grok 4.1 Fast xAI | 84 | $10.0 | 8.40 |
| DeepSeek V3.2 DeepSeek | 85 | $11.8 | 7.23 |
| Qwen3 235B Alibaba (Qwen) | 84 | $13.6 | 6.16 |
| Llama 4 Maverick Meta | 78 | $14.9 | 5.23 |
| GPT-5 mini OpenAI | 83 | $23.5 | 3.53 |
Value = quality points per dollar of monthly spend at your volume. Quality scores are benchmark-informed estimates, not evals — validate on your own workload before migrating.
Price your own agents & tool use volume
These rankings use a typical monthly volume. Plug in your real token counts and the free estimator re-ranks every model on cost and quality.
Open the estimatorEmbed the live pick
A self-updating SVG badge of the balanced pick for agents & tool use. Drop it in a README or blog — it re-renders the day the catalog changes.
Best models for other tasks
Prices are list $/1M tokens, hand-maintained from public provider pricing. Quality scores are benchmark-informed 0-100 estimates, not evals — validate on your own workload before moving production traffic. Not affiliated with any model provider. Compare Claude Opus 4.5 vs Kimi K2