Cloud Providers

This page covers setup for cloud-based API providers. For GitHub Copilot, see GitHub Copilot. For ChatGPT OAuth, see ChatGPT OAuth Login.

Anthropic

Models: Claude Fable 5.1/5, Opus 5.5/5/4.8, Sonnet 5.5/5, Haiku 5.5, and active 4.x models.

1. Get an API key from console.anthropic.com.
2. Run fermi init and select Anthropic (Claude).
3. Paste your API key when prompted.

The key is stored as ANTHROPIC_API_KEY in ~/.fermi/.env.

Context and thinking

Opus/Sonnet 4.6 now have GA 1M context without a beta header. New 5.x models also support 1M. Legacy 1M selectors remain compatible.

Opus 5.5/Fable always think, and Sonnet 5.5 cannot fully disable thinking. They expose low/medium/high/xhigh/max. Haiku 5.5 also supports off.

OpenAI

Models: GPT-6.1 Sol, GPT-6 Astra/Sol/Luna, GPT-5.6 Sol/Terra/Luna, GPT-5.5/5.4/5.2

1. Get an API key from platform.openai.com.
2. Run fermi init and select OpenAI.
3. Paste your API key when prompted.

The key is stored as OPENAI_API_KEY in ~/.fermi/.env.

Alternatively, use your ChatGPT account via OAuth. See ChatGPT OAuth Login.

Thinking levels: GPT-5.6 and GPT-6 Sol/Luna: none, low, medium, high, xhigh, max. Astra and 6.1 Sol: low through max. Ultra is not a native API effort.

DeepSeek

Models: DeepSeek V4.1 Flash (deepseek-flash, vision) and V4 Pro (0813, text). The legacy V4 Flash ID redirects to V4.1.

1. Get an API key from DeepSeek’s developer portal.
2. Run fermi init and select DeepSeek.
3. Paste your API key when prompted.

The key is stored as a managed slot in ~/.fermi/.env.

Kimi / Moonshot

Models: Public API: Kimi K3, K2.7 Code/Highspeed, K2.6. Coding: k3, k3-256k, kimi-for-coding (K2.8 Preview), kimi-for-coding-highspeed.

Available through three endpoint variants:

Variant	Endpoint	Fermi Slot	Detected External Env
Kimi-Global	api.moonshot.ai	FERMI_KIMI_API_KEY	MOONSHOT_API_KEY, KIMI_API_KEY
Kimi-China	api.moonshot.cn	FERMI_KIMI_CN_API_KEY	MOONSHOT_API_KEY, KIMI_CN_API_KEY
Kimi-Code	api.kimi.com/coding	FERMI_KIMI_CODE_API_KEY	KIMI_CODE_API_KEY

Caution

The Kimi-Code endpoint is restricted to whitelisted agents. You may receive a 403 error. Use kimi or kimi-cn (standard API) instead.

1. Get an API key from Moonshot’s developer portal.
2. Run fermi init and select Moonshot (Kimi), then pick your variant.
3. Import a detected env var or paste your API key.

GLM / Zhipu

Models: GLM-5.3, 5.3 Flash, 5.2, and compatible older models. 5.3 FlashX is API-only. 5.3 supports low/high/max and always thinks.

Available through four endpoint variants:

Variant	Endpoint	Fermi Slot	Detected External Env
GLM-China	open.bigmodel.cn	FERMI_GLM_API_KEY	GLM_API_KEY
GLM-Global	api.z.ai	FERMI_GLM_INTL_API_KEY	GLM_INTL_API_KEY
GLM-China-Code	open.bigmodel.cn/api/coding	FERMI_GLM_CODE_API_KEY	GLM_CODE_API_KEY
GLM-Global-Code	api.z.ai/api/coding	FERMI_GLM_INTL_CODE_API_KEY	GLM_INTL_CODE_API_KEY

Caution

The GLM coding endpoints are restricted to whitelisted agents. Use glm or glm-intl (standard API) instead.

1. Get an API key from Zhipu’s developer portal.
2. Run fermi init and select z.ai (GLM/Zhipu), then pick your variant.
3. Import a detected env var or paste your API key.

MiniMax

Models: MiniMax M3.1 Flash Preview (M Plan only), M3, M2.7/2.5 and Highspeed.

Available through two endpoint variants:

Variant	Endpoint	Fermi Slot	Detected External Env
MiniMax-Global	api.minimax.io	FERMI_MINIMAX_API_KEY	MINIMAX_API_KEY
MiniMax-China	api.minimaxi.com	FERMI_MINIMAX_CN_API_KEY	MINIMAX_CN_API_KEY

1. Get an API key from MiniMax’s developer portal.
2. Run fermi init and select MiniMax, then pick your variant.
3. Import a detected env var or paste your API key.

Xiaomi (MiMo)

Models: MiMo V2.6 Pro, Flash, UltraSpeed (access required); V2.5/Pro retire on 2026-10-21.

1. Get an API key from Xiaomi’s developer portal.
2. Run fermi init and select Xiaomi (MiMo).
3. Paste your API key when prompted.

The key is stored as a managed slot in ~/.fermi/.env.

Qwen / DashScope

Models: Qwen3.8 Max, Max 0902, Flash, Omni Flash, 3.7 Plus/Flash; CN/Intl also offer 3.8 27B and 2.4T A95B. Compatible older models remain.

Available through three regional endpoint variants:

Variant	Region	Fermi Slot	Detected External Env
Qwen	China (Beijing)	FERMI_QWEN_API_KEY	DASHSCOPE_API_KEY, QWEN_API_KEY
Qwen-Intl	Singapore	FERMI_QWEN_INTL_API_KEY	DASHSCOPE_INTL_API_KEY, QWEN_INTL_API_KEY
Qwen-US	United States	FERMI_QWEN_US_API_KEY	DASHSCOPE_US_API_KEY, QWEN_US_API_KEY

1. Get an API key from Alibaba Cloud’s DashScope console.
2. Run fermi init and select Qwen / DashScope, then pick your region.
3. Import a detected env var or paste your API key.

Responses uses the new /compatible-mode/v1 path. Native context is 1M, but the API reserves about 20% and may silently truncate excess input, so Fermi presets budget 800k. 3.8 exposes none/low/medium/xhigh; the 2.4T open model is thinking-only.

OpenRouter

Models: Multi-vendor curated presets (Claude, GPT, Kimi, MiniMax, GLM, DeepSeek, Qwen, Xiaomi) plus any custom model.

OpenRouter acts as a unified API gateway to multiple model providers.

1. Get an API key from openrouter.ai.
2. Run fermi init and select OpenRouter.
3. Paste your API key.
4. Pick from the curated model presets.

The key is stored as OPENROUTER_API_KEY in ~/.fermi/.env.