AI model prices for coding agents
What each model costs through its API, how much context it takes and which AI agents can run it. Prices are per million tokens in US dollars, from LiteLLM's open price list, refreshed daily; each model page links to the provider's own pricing page.
| Model | Input | Cached | Output | Example task | Context | Agents |
|---|---|---|---|---|---|---|
Claude Fable 5.1Anthropic · frontier |
$10 | $0.25 | $50 | $15 | 1M | 78 |
GPT-6 AstraOpenAI · frontier |
$10 | $1 | $50 | $15 | 922K | 74 |
Claude Opus 5.5Anthropic · frontier |
$4 | $0.20 | $20 | $6 | 1M | 78 |
Kimi K3Moonshot AI · frontier |
$3 | $0.30 | $15 | $4.5 | 1M | 56 |
GPT-5.6 TerraOpenAI · mid |
$2 | $0.20 | $12 | $3.2 | 922K | 74 |
Gemini 3.1 Pro (preview)Google · mid |
$2 | $0.20 | $12 | $3.2 | 1M | 77 |
Claude Sonnet 5.5Anthropic · mid |
$2 | $0.10 | $10 | $3 | 1M | 78 |
GPT-6.1 SolOpenAI · mid |
$2 | $0.10 | $10 | $3 | 922K | 74 |
Mistral Medium 3.5Mistral AI · mid |
$1.5 | $0.15 | $7.5 | $2.25 | 262K | 55 |
Grok 4.7xAI · mid |
$2 | $0.50 | $6 | $2.6 | 500K | 60 |
Qwen3.8 MaxAlibaba (Qwen) · mid |
$2 | $0.25 | $6 | $2.6 | 992K | 58 |
GLM-5.3Z.ai · mid |
$1.4 | $0.26 | $4.4 | $1.84 | 1M | 51 |
Kimi K2.7 CodeMoonshot AI · mid |
$0.95 | $0.19 | $4 | $1.35 | 262K | 56 |
DeepSeek V4 ProDeepSeek · mid |
$1.32 | $0.04 | $3.96 | $1.72 | 1M | 60 |
Gemini 3.8 FlashGoogle · mid |
$0.75 | $0.07 | $3.75 | $1.12 | 1M | 77 |
Grok Code Fast 1xAI · fast |
$1 | $0.20 | $2 | $1.2 | 256K | 60 |
DeepSeek V4 FlashDeepSeek · fast |
$0.30 | $0.01 | $1.2 | $0.42 | 1M | 60 |
MiniMax M3MiniMax · fast |
$0.30 | $0.06 | $1.2 | $0.42 | 1M | 51 |
MiMo V2.6 ProXiaomi (MiMo) · fast |
$0.43 | $0.00 | $0.87 | $0.52 | 1M | 51 |
Mistral Small (2603)Mistral AI · fast |
$0.15 | $0.01 | $0.60 | $0.21 | 262K | 55 |
Claude Haiku 5.5Anthropic · fast |
$0.10 | $0.01 | $0.50 | $0.15 | 1M | 78 |
GLM-5.3 FlashZ.ai · fast |
$0.15 | $0.03 | $0.50 | $0.20 | 1M | 51 |
GPT-6 LunaOpenAI · fast |
$0.10 | $0.01 | $0.50 | $0.15 | 922K | 74 |
Qwen3.8 FlashAlibaba (Qwen) · fast |
$0.15 | $0.02 | $0.47 | $0.20 | 992K | 58 |
MiMo V2.6 FlashXiaomi (MiMo) · fast |
$0.14 | $0.00 | $0.28 | $0.17 | 1M | 51 |
"Example task" is a rough yardstick: 1 million input tokens plus 100,000 output tokens, about one long agent session on a mid-size codebase, without prompt caching. Coding agents re-send context on every step, so cached-input prices matter: most agents cache automatically. Tiers: frontier = $15+ per million output tokens, mid = $3-15, fast = under $3.
How people combine models
A common setup is a strong model for planning and review and a fast, cheap one for the edits in between. In Claude Code that can be a frontier model in plan mode and a smaller model for subagents; in agents that take any provider (OpenCode, Cline, Aider and others) you pick a model per role. Compare the "Example task" column within each tier to see what the switch saves, and check each model page for the agents that support it.
Subscription plans (Claude Pro and Max, ChatGPT plans, Cursor, Copilot) bill differently: you pay a monthly fee with usage limits instead of per token. See agent pricing for those.