Docs/Getting Started/Models

Models

Requests run through the codegraff gateway, which proxies these model aliases to their upstream providers. Prices are per million tokens (USD).

Available models

ModelProviderContextInputOutputCached in
gemini-3.7-flashGoogle1M$0.75$3.75$0.075
claude-opus-4.8Anthropic1M$5.00$25.00$0.50
gpt-5.5OpenAI400K$5.00$30.00$0.50
claude-sonnet-4.6Anthropic1M$3.00$15.00$0.30
muse-spark-1.2Meta1M$1.25$4.25$0.15
muse-spark-1.2-contributorMeta (Contributor)1M$0.10$0.20$0.002
deepseek-v4-proDeepSeek1M$0.435$0.87$0.0036
deepseek-v4-flashDeepSeek1M$0.14$0.28$0.0028
deepseek-v4-flash-fastDeepSeek1M$0.294$0.588$0.0735
grok-buildxAI256K$1.00$2.00$0.20
kimi-k3Moonshot1M$3.60$18.00$0.36
kimi-k2.7-codeMoonshot256K$1.14$4.80$0.23
kimi-k2.7-code-highspeedMoonshot1M$2.28$9.60$0.46
minimax-m3MiniMax1M$0.72$2.88$0.144
glm-5.2Zhipu128K$1.68$5.28$0.312
mimo-v2.5-pro-ultraspeedXiaomi1M$1.305$2.61$0.0108
mimo-v2.5-proXiaomi1M$0.435$0.87$0.0036
stealth-ox-alphaStealth1MFreeFreeFree
mimo-v2.5Xiaomi1M$0.14$0.28$0.0028

Muse Spark contributor data use

Requests sent to muse-spark-1.2-contributor may be used by Meta to train future models. Use muse-spark-1.2 when prompts and completions must not be used for Meta model training.

Prompt caching

Cached input tokens are billed at the lower cached-input rate shown above. The gateway meters cache hits automatically for every provider, including Claude and Gemini, so repeated context in multi-turn and tool-calling loops is discounted with no code changes. Gemini 3.7 Flash uses the Interactions API with implicit caching (minimum 4,096 input tokens). Pass previous_interaction_id to continue a stored thread. Its built-in Google Search, Maps, computer-use, file-search, code-execution, and URL-context tools are blocked — they bill outside the token meter. Your own function tools still work. Prices above are introductory through 2026-12-31.

Using a model

Pass any alias as --model on the CLI, as the model option in the SDKs, or as the model field of an OpenAI-compatible request. The SDKs also accept a per-turn model override.

Color theme
Theme follows your system