Skip to main content

Model Configuration

Cody CLI ships with 9 top coding models via the AINative gateway. All models support tool use (file read/write, bash, search, agents).

Available Models

Model IDProviderTierContextBenchmarkPricing
deepseek-4-flashDigitalOceanFree131K79% SWE-BenchFree
kimi-k2DigitalOceanFree131K80.2% SWE-BenchFree
qwen3-coder-flashDigitalOceanFree131KFree
llama-3.3-70b-instructMetaFree131KFree
qwen-coder-32bHugging FaceProfessional131K$0.28/$1.10 per MTok
glm-5NVIDIAProfessional131KUsage-based
kimi-k2-thinkingNVIDIAProfessional131KUsage-based
claude-sonnet-4.5BYOKEnterprise200K$3/$15 per MTok
claude-opus-4BYOKEnterprise200K$15/$75 per MTok

Defaults

  • Free tier: deepseek-4-flash — best free coding model (79% SWE-Bench)
  • Paid tier: kimi-k2 — best open-source coding model (80.2% SWE-Bench)

BYOK Models

claude-sonnet-4.5 and claude-opus-4 require your own API key (Bring Your Own Key). These are routed through the AINative gateway but billed at the upstream provider's rates.

Selecting a Model

Command-line flag

cody --model kimi-k2

Interactive model picker

Run /model inside an interactive session to switch models on the fly. Models are filtered by your subscription tier.

Inline config

/config model=kimi-k2

Settings file

Add to ~/.cody/settings.json:

{
"model": "kimi-k2"
}

Context Windows

Each model has a maximum context window. Cody automatically manages context — when a conversation approaches the limit, older turns are summarized to stay within bounds. You don't need to manage this manually.

Context SizeModels
131K tokensdeepseek-4-flash, kimi-k2, qwen3-coder-flash, llama-3.3-70b-instruct, qwen-coder-32b, glm-5, kimi-k2-thinking
200K tokensclaude-sonnet-4.5, claude-opus-4

Tool Support

All 9 models in the registry support tool use:

  • File read, write, edit
  • Bash command execution
  • Grep, glob, web search
  • Agent spawning (up to 5 levels)
  • MCP server tools

Fast Mode

Toggle fast mode with /fast in a session, or launch with:

cody --fast

Fast mode uses the same model with faster output — it does not downgrade to a smaller model.

Model Registry

The CLI maintains a static fallback registry of 9 models. On startup, the live catalog from api.ainative.studio overrides the static list — so new models appear automatically as they're added to the platform, without a CLI update.