Models reference
The model lineup across Chat and Agent Dev, what the reasoning-effort levels trade off, and what each surface uses by default.
The two surfaces source models differently. Chat models are hosted and appear in the composer after you sign in to Wingman. Build (formerly Agent Dev) starts on the same hosted Axelera AI models, run on your device by the Wingman engine with your Wingman sign-in, and can switch to an installed, signed-in Claude Code, Codex, or Kimi CLI. Its model picker reflects the selected provider and its sign-in state.
Chat models
Chat defaults to Axelera Earth 1.0, the deepest Axelera AI model, so every new conversation starts with the strongest answer. Pick a model from the model menu and Chat keeps it. A model you did not pick yourself in the menu, including the Axelera Venus 1.0 default from earlier releases, now reads as Axelera Earth 1.0.
| Family | Current picks | Good for |
|---|---|---|
| Axelera AI | axelera-earth-1.0 (default), axelera-venus-1.0, axelera-mercury-1.0 | Earth for the deepest multi-step work, Venus for balance, Mercury for the fastest answers. All three know the Voyager SDK best. |
| Claude | Fable 5, Opus 4.8 / 4.7 / 4.6, Sonnet 4.6, Haiku 4.5 | Deep reasoning and long documents. Haiku is the fast, light option. |
| GPT | GPT-5.5, GPT-5.5 Pro, the GPT-5.4 family, GPT-5.2, GPT-5.1 | Strong general-purpose family, with mini and nano tiers for speed. |
| Gemini | Gemini 3.1 Pro Preview, Gemini 3 Flash Preview | Multimodal input and analysis. |
| DeepSeek | V3.2, V3.2 Thinking | Cost-efficient reasoning. |
| Grok | Grok 4.20 Beta, Grok 4 | xAI’s line, including a non-reasoning 4.20 variant. |
| Kimi | K2.5, K2, K2 Thinking | Long-context work with thinking and turbo variants. |
| MiniMax | M2.7, M2.5, plus Highspeed variants | Fast agentic workloads. |
| GLM | GLM-5 | Zhipu’s flagship generalist. |
| MiMo | MiMo-V2 Pro / Omni / Flash | Xiaomi’s multimodal line. |
Agent Dev providers
Build offers Axelera AI, which it starts on, and the Claude Code, Codex, and Kimi CLIs you install and sign in to. The provider, model, and effort controls sit one row below the Build composer. The label starts with the Requested model and changes to Running only when the active provider reports the model it observed.
| Provider | Models in the picker | Runs through |
|---|---|---|
| Axelera AI | Axelera Mercury 1.0, Axelera Venus 1.0, Axelera Earth 1.0 | The Wingman engine on your device, with your Wingman sign-in. Venus and Earth draw on your credits. When your credits are used up, your seat reaches its monthly cap or an internal account’s daily allowance is used, the chat continues on Mercury 1.0 and a note above the composer says so. A turn can also run on Mercury without a note while your other chats hold the credits that are left, or when one turn alone is too long to fund. |
| Claude | fable, opus, sonnet | Your installed, signed-in Claude Code CLI. |
| OpenAI | gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna | Your installed, signed-in Codex CLI. The picker always lists the newest GPT family your CLI reports. |
| Kimi | kimi-code/k3, kimi-code/kimi-for-coding, kimi-code/kimi-for-coding-highspeed | Your installed, signed-in Kimi CLI. |
Reasoning effort levels
Reasoning effort controls how long the model thinks before it answers. Higher effort helps on hard, multi-step problems; it also costs time. Drop the effort for quick edits and small questions. In Chat, on a model you run with your own key, the effort control shows the level the model runs at until you pick one: xhigh on GPT-6.1 Sol and GPT-6 Astra, medium on Claude Opus 5.5 and high on the other Claude models.
| Provider and model | Available effort levels | Highest tier |
|---|---|---|
| Claude Code | low, medium, high, xhigh, max | Max; Fable 5 labels the same top stop Ultracode. |
| Codex | low (shown as Light), medium, high, xhigh, ultra | Ultra. |
| Kimi K3 | low, high, max | Max. Kimi K3 supports Low, High, and Max. |
Defaults by surface
What each surface uses out of the box.
| Surface | Default | Notes |
|---|---|---|
| Chat | Axelera Earth 1.0 (axelera-earth-1.0) | Change it per conversation from the composer’s model menu. Chat keeps a model you pick there. |
| Chat background services | Axelera Venus 1.0 and Axelera Auto | Venus 1.0 summarises history and names agents; Axelera Auto names topics, translates and rewrites. Change them in Settings, under Service Model. |
| Build | Axelera Earth 1.0 (axelera-earth-1.0) | With approval policy on-request and sandbox workspace-write. An install that saved another provider or another Axelera AI model before this default moves here once, and a previous model that Build still offers stays one click away in its menu; a local model stays as it is. Any install in Full access goes to Ask once. A model or provider you pick afterwards stays picked. |
| Build, Claude provider | Fable 5.1 (claude-fable-5-1) at high effort | Applied when you switch the provider to Claude. |