Model Catalog
v0.7.0-alpha snapshot. Model ids move fast. This page tracks the curated ids shipped in the setup wizard as of the current release. The live source of truth is
curated_models_for_provider/default_model_for_providerinsrc/onboard/wizard.rs; the/modelpicker in the TUI reads the same lists.
Model ids are free-form strings — the runtime passes whatever you set in
default_model straight to the provider. Nothing gates them, so any id the
provider accepts works even if it is not listed here. The curated ids below are
just the vetted defaults the onboarding wizard and /model picker offer.
How the per-provider default works
When you run the setup wizard and accept the suggested model for a provider, the
wizard writes that provider’s default into default_model (see
default_model_for_provider in src/onboard/wizard.rs). This is a
wizard-time default — the id it pre-fills for the chosen provider — not a
live runtime fallback.
If default_model is left unset entirely in config.toml, the agent loop uses
a hardcoded fallback of anthropic/claude-sonnet-4 rather than the per-provider
default (see src/agent/loop_.rs). In practice default_model is always
populated after onboarding, so the per-provider default is what you get.
Recommended default per provider
Canonical provider name (aliases resolve to these) and the id
default_model_for_provider returns:
| Provider | Default model id |
|---|---|
openrouter | anthropic/claude-sonnet-4.6 |
anthropic | claude-sonnet-4-6 |
openai | gpt-5.5 |
openai-codex | gpt-5.3-codex |
gemini (google) | gemini-3-pro |
deepseek | deepseek-v4-pro |
xai (grok) | grok-4-1-fast-reasoning |
qwen | qwen3.7-plus |
qwen-code | qwen3.6-coder-plus |
moonshot (kimi) | kimi-k2.6 |
kimi-code | kimi-for-coding |
glm / zai | glm-5.2 |
minimax | MiniMax-M3 |
mistral | mistral-large-latest |
cohere | command-a-03-2025 |
perplexity | sonar-pro |
groq | llama-spark |
venice | zai-org-glm-5.1 |
fireworks | accounts/fireworks/models/llama-spark |
together-ai | meta-llama/Llama-Spark |
nvidia | meta/llama-spark |
ollama | llama-spark |
llamacpp | ggml-org/llama-spark-GGUF |
bedrock | anthropic.claude-sonnet-4-6-v1:0 |
Curated menus (flagships)
The wizard and /model picker present these per-provider menus. The entry
marked “(default)” is what default_model_for_provider returns — note that
Moonshot and Qwen list the newest id first while still defaulting to the stable
flagship.
Anthropic
| Model id | Notes |
|---|---|
claude-sonnet-4-6 | Balanced, recommended (default) |
claude-opus-4-8 | Best quality |
claude-opus-4-7 | Previous flagship |
claude-haiku-4-5-20251001 | Fastest, cheapest |
claude-sonnet-4-5-20250929 | Legacy |
OpenAI
| Model id | Notes |
|---|---|
gpt-5.5 | Latest flagship, recommended (default) |
gpt-5.4 | Previous flagship, stable |
gpt-5.3-codex | Agentic coding |
gpt-5.2 | Older, stable |
gpt-5-mini | Faster, cheaper |
gpt-5-nano | Lowest latency/cost |
o4-mini | Reasoning fallback |
Removed phantom ids (0.6.80). The
openailist no longer includesgpt-5.5-codex, and the baregpt-5.3id is gone — neither exists on the OpenAI API. Theopenai-codexprovider now defaults togpt-5.3-codex(the newest codex id that actually exists), and regression tests assert the phantom ids never reappear.
OpenAI Codex (openai-codex)
| Model id | Notes |
|---|---|
gpt-5.3-codex | Latest, recommended (default) |
gpt-5.2-codex | Older flagship |
gpt-5.1-codex-mini | Cheap, capable |
gpt-5-codex | Legacy, stable |
o4-mini | Reasoning fallback |
DeepSeek
| Model id | Notes |
|---|---|
deepseek-v4-pro | Latest flagship, recommended (default) |
deepseek-v4-flash | Fast, cost-efficient |
deepseek-chat | Mapped to V3.2 non-thinking |
deepseek-reasoner | Mapped to V3.2 thinking |
Qwen
| Model id | Notes |
|---|---|
qwen3.7-max | Latest flagship, highest quality |
qwen3.7-plus | Balanced (default) |
qwen3.6-max | Previous flagship |
qwen3.6-plus | Balanced |
qwen3.6-turbo | Fast, cost-efficient |
GLM / Z.ai (glm, zai)
| Model id | Notes |
|---|---|
glm-5.2 | Latest flagship, recommended (default) |
glm-5.1 | Previous flagship |
glm-5 | Stable |
glm-4.7 | Strong general-purpose, stable |
glm-4.5-air | Lower latency |
Moonshot / Kimi (moonshot)
| Model id | Notes |
|---|---|
kimi-k2.7-code | Latest, agentic coding |
kimi-k2.6 | Flagship, recommended (default) |
kimi-k2.5 | Previous flagship |
kimi-k2-thinking | Deep reasoning + tool use |
kimi-k2-0905-preview | Legacy |
MiniMax
| Model id | Notes |
|---|---|
MiniMax-M3 | Latest flagship, recommended (default) |
MiniMax-M2.7 | Previous flagship |
MiniMax-M2.7-highspeed | Fast |
MiniMax-M2.5 | Stable |
MiniMax-M2.1 | Strong coding/reasoning, stable |
xAI / Grok (xai)
| Model id | Notes |
|---|---|
grok-4-1-fast-reasoning | Recommended (default) |
grok-4-1-fast-non-reasoning | Low latency |
grok-code-fast-1 | Coding specialist |
grok-4 | Max quality |
Mistral
| Model id | Notes |
|---|---|
mistral-large-latest | Latest flagship (default) |
mistral-medium-latest | Balanced |
codestral-latest | Code-focused |
devstral-latest | Software-engineering specialist |
Cohere
| Model id | Notes |
|---|---|
command-a-03-2025 | Flagship enterprise model (default) |
command-a-reasoning-08-2025 | Agentic reasoning |
command-r-08-2024 | Stable fast baseline |
Gemini
| Model id | Notes |
|---|---|
gemini-3-pro | Latest frontier reasoning, recommended (default) |
gemini-3-flash | Best price/performance |
gemini-2.5-pro | Previous flagship, stable |
gemini-2.5-flash | Legacy fast tier |
gemini-2.5-flash-lite | Lowest cost |
OpenRouter and aggregators
openrouter (and astrai) use slash-prefixed, cross-provider ids because they
route to many upstreams. The curated OpenRouter menu leads with
anthropic/claude-sonnet-4.6 (the default), and also offers
anthropic/claude-opus-4.7, openai/gpt-5.5, google/gemini-3-pro,
x-ai/grok-4.1-fast, deepseek/deepseek-v3.2, moonshotai/kimi-k2.6, and
z-ai/glm-5.1, among others. Set any provider-prefixed id the aggregator
supports.
Overriding at runtime
For most providers the wizard can also fetch the live model list from the
provider’s /models endpoint (supports_live_model_fetch in
src/onboard/wizard.rs), so the picker shows current ids beyond this curated
set. You can always set default_model by hand or switch mid-session with
/model. See Providers for provider slugs,
routing, and fallbacks.