Best models for Claude Code
The best model for Claude Code right now is GPT-5.6 Sol, with a 77.4 Artificial Analysis Coding Index; GPT-5.6 Terra (76.7) and Claude Fable 5 (76.5) follow. Claude Code runs long agentic sessions with heavy tool use — the Agentic and Coding indexes matter most, and prompt caching sharply cuts session cost.
Ranked by the Artificial Analysis Coding Index, filtered to models with the tool-calling support agentic editors require. Speed, context and price are shown per row — claude Code runs long agentic sessions with heavy tool use — the Agentic and Coding indexes matter most, and prompt caching sharply cuts session cost.
- 1
gpt-5.6-solReasoningToolsJSON+1$5.00/M · 1.1M ctx77.4Coding - 2gpt-5.6-terraReasoningToolsJSON+1$2.50/M · 1.1M ctx76.7Coding
- 3
claude-fable-5ReasoningToolsJSON+1$10.00/M · 1M ctx76.5Coding - 4
kimi-k3ReasoningToolsJSON+1$3.00/M · 1.0M ctx76.2Coding - 5gpt-5.5ReasoningToolsJSON+1$5.00/M · 1.1M ctx74.9Coding
- 6claude-opus-4.8ReasoningToolsJSON+1$5.00/M · 1M ctx74.3Coding
- 7claude-opus-4.7ReasoningToolsJSON+1$5.00/M · 1M ctx73.6Coding
- 8
grok-4.5ReasoningToolsJSON+1$2.00/M · 500K ctx72.4Coding - 9claude-sonnet-5ReasoningToolsJSON+1$2.00/M · 1M ctx71.5Coding
- 10gpt-5.6-lunaReasoningToolsJSON+1$1.00/M · 1.1M ctx71.4Coding
- 11Mmuse-spark-1.1ReasoningToolsJSON+2$1.25/M · 1.0M ctx71.3Coding
- 12gpt-5.4ReasoningToolsJSON+1$2.50/M · 1.1M ctx71.1Coding
- 13
gemini-3.5-flashReasoningToolsJSON+2$1.50/M · 1.0M ctx70.1Coding - 14gemini-3.6-flashReasoningToolsJSON+2$1.50/M · 1.0M ctx69.2Coding
- 15
glm-5.2ReasoningToolsJSON$0.818/M · 1.0M ctx68.8Coding - 16gemini-3.1-pro-previewReasoningToolsJSON+2$2.00/M · 1.0M ctx68.8Coding
- 17
qwen3.7-maxReasoningToolsJSON$1.48/M · 1M ctx66.0Coding - 18kimi-k2.6ReasoningToolsJSON+1$0.684/M · 262K ctx61.8Coding
- 19kimi-k2.7-codeReasoningToolsJSON+1$0.710/M · 262K ctx60.8Coding
- 20
mimo-v2.5-proReasoningToolsJSON$0.435/M · 1.1M ctx60.2Coding
Frequently asked
What is the best model for Claude Code?
GPT-5.6 Sol leads on the Coding Index (77.4) among models with the tool-calling support Claude Code requires. Claude Code runs long agentic sessions with heavy tool use — the Agentic and Coding indexes matter most, and prompt caching sharply cuts session cost.
What's a cheaper model to use in Claude Code?
GPT-5.6 Terra (76.7 coding, $2.50/M) and Kimi K3 (76.2 coding, $3.00/M) give most of the leader's capability at a lower per-token price.