Best LLMs for Tool Calling — Anthropic
The best Anthropic model for tool calling is Claude Fable 5, completing 98.5% of Tau²-Bench's multi-turn tool-use tasks. Claude Opus 4.8 (94.4%) and Claude Opus 4.7 (88.6%) round out the top three.
- 1
claude-fable-5ReasoningToolsJSON+159.9 intel · $10.00/M · 82 t/s98.5%τ²-Bench - 2claude-opus-4.8ReasoningToolsJSON+155.7 intel · $5.00/M · 1M ctx94.4%τ²-Bench
- 3claude-opus-4.7ReasoningToolsJSON+153.5 intel · $5.00/M · 1M ctx88.6%τ²-Bench
- 4claude-opus-4.5ReasoningToolsJSON+134.7 intel · $5.00/M · 52 t/s86.3%τ²-Bench
- 5claude-opus-4.6ReasoningToolsJSON+137.8 intel · $5.00/M · 1M ctx84.8%τ²-Bench
- 6claude-sonnet-4.6ReasoningToolsJSON+135.9 intel · $3.00/M · 1M ctx79.5%τ²-Bench
- 7claude-sonnet-4.5ReasoningToolsJSON+129.3 intel · $3.00/M · 36 t/s70.5%τ²-Bench
- 8claude-sonnet-4ReasoningToolsVision25.5 intel · $3.00/M · 46 t/s52.3%τ²-Bench
- 9claude-haiku-4.5ReasoningToolsJSON+123.7 intel · $1.00/M · 91 t/s32.5%τ²-Bench
- 10claude-opus-5ReasoningToolsJSON+160.7 intel · $5.00/M · 60 t/s30.3%τ²-Bench
- 11claude-sonnet-5ReasoningToolsJSON+153.4 intel · $2.00/M · 95 t/s28.2%τ²-Bench
- 12claude-3-haikuToolsVision3.9 intel · $0.250/M · 200K ctx21.1%τ²-Bench
How this is ranked
AI models ranked by Tau²-Bench — multi-turn conversations where the model has to call the right tools, in the right order, against a real API to complete a customer task. This measures whether function calling actually works under pressure, which is a different question from whether a model supports the parameter at all.
Frequently asked
Which Anthropic model is best at tool calling?
The best Anthropic model for tool calling is Claude Fable 5, completing 98.5% of Tau²-Bench's multi-turn tool-use tasks. Claude Opus 4.8 (94.4%) and Claude Opus 4.7 (88.6%) round out the top three.
What's a good alternative to Claude Fable 5?
Claude Opus 4.8 (94.4%) is the closest alternative on this metric, followed by Claude Opus 4.7 (88.6%). See the full ranking above for the tradeoffs.
How many Anthropic models are there?
modelgrep tracks 17 Anthropic models with live benchmarks, speed, latency and per-provider pricing, led on intelligence by Claude Opus 5. 12 of them qualify for this ranking.