MiMo V2.5 has a 1-million-token context window, but its direct public API uses Chat Completions. Current Codex custom providers use the Responses protocol. Requesty supplies the missing translation, which makes MiMo usable without an unsupported Codex setting.
This is the Windows config and the one gotcha. For Codex’s general provider model, see Codex CLI custom provider setup.
Step 1: Install Codex and create a Requesty key
Install Codex CLI, then create a Requesty API key. On Windows, you can use PowerShell or WSL. The Codex CLI on Windows with WSL guide covers the WSL setup.
Step 2: Add MiMo to config.toml
model = "xiaomi/mimo-v2.5"
model_provider = "requesty"
model_supports_reasoning_summaries = false
[model_providers.requesty]
name = "Requesty"
base_url = "https://router.requesty.ai/v1"
env_key = "REQUESTY_API_KEY"
wire_api = "responses"Export the key and run:
export REQUESTY_API_KEY="your-requesty-key"
codexPowerShell uses $env:REQUESTY_API_KEY = "your-requesty-key" before codex.
Step 3: Use the large context carefully
MiMo accepts up to 1 million tokens of context. That capacity is useful for large repositories, although sending more files still increases token usage. In WSL, keep active projects in the Linux filesystem for faster file operations. See access your Windows files from WSL.
Current Requesty model IDs
| xiaomi/mimo-v2.5 | Standard MiMo V2.5; 1M context |
|---|---|
| xiaomi/mimo-v2.5-pro | Pro variant; check the live catalog before switching |
Pricing
Requesty’s catalog listed standard MiMo V2.5 at $0.14 per million input tokens, $0.28 per million output tokens, and $0.0028 per million cached input tokens on August 24, 2026. Provider rates can change, so check the live MiMo V2.5 listing before estimating a long session. See Xiaomi MiMo pricing and free tiers for the direct-provider options.
Troubleshooting
- Format or streaming errors: confirm
wire_api = "responses"and update Codex. - Auth errors: confirm
REQUESTY_API_KEYis set in the shell that launches Codex. - “Model not found”: use the provider-prefixed ID
xiaomi/mimo-v2.5. - Requesty 400 errors: keep
model_supports_reasoning_summaries = falseat the top level. - Changes ignored, restart Codex;
config.tomlloads at startup.
MiMo + Codex checklist
- Codex CLI installed and Requesty key created
- Requesty provider points to https://router.requesty.ai/v1
- wire_api is set to responses
- model is set to xiaomi/mimo-v2.5
- REQUESTY_API_KEY is exported in the current shell
Verify the result
Run a small edit and inspect the diff. Then check Requesty’s activity log to confirm the request used xiaomi/mimo-v2.5. This verifies the key, model ID, Responses translation, and basic tool use before you load a large repository.
To run MiMo with Claude Code, see run Xiaomi MiMo V2.5 Pro with Claude Code.
