Use DeepSeek-V4-Flash-Vision-Exp in Kimi Code
Add the text models available to your API key to Kimi's model picker, alongside your existing Kimi account.
Selected model: DeepSeek-V4-Flash-Vision-Exp. The copyable prompt below is already customized for it.
Verified October 22, 2026 · Kimi Code 0.29.1 · Requires Node.js 20+, Windows PowerShell 5.1+ and an API key.
1. Copy this prompt into Codex
Let Codex inspect the installer and configure your local client. Enter your API key only at the local hidden prompt. The setup fetches your account's current model list; the public catalog may differ from your account.
Add ChineseModels text/chat models to my Kimi Code model picker. Guide: https://chinesemodels.ai/documentation/kimi-code Installer: https://chinesemodels.ai/downloads/chinesemodels-kimi-windows.zip API base: https://chinesemodels.ai/v1 Preferred model after installation: DeepSeek-V4-Flash-Vision-Exp Keep all available text models selectable. 1. Detect my OS, Kimi version, and KIMI_CODE_HOME. This installer targets Windows and Kimi Code 0.29.1 (~/.kimi-code), not legacy kimi-cli (~/.kimi). 2. Back up config.toml. Preserve my existing Kimi K3 login, providers, default model, hooks, and unrelated settings. 3. Download and inspect the installer. Ask for my API key through its hidden local prompt; never print it or put it in source code, a browser URL, or this conversation. Kimi's native api_key field stores the credential in the local config as plain text. Offer api_key_env if I prefer to manage an environment variable. 4. Run install-kimi-windows.ps1. It fetches my authenticated /v1/models list and adds text/chat models under chinesemodels/<exact model ID>. Exclude image/video generators, including incorrectly tagged entries. 5. Use the native openai provider with the /v1 base URL and reasoning_content support. Thinking metadata is model-specific. Native reasoning_effort does not guarantee vendor-specific thinking.type or enable_thinking controls; verify the wire parameters before promising thinking on/off. 6. Run kimi doctor config. Test a short reply from chinesemodels/DeepSeek-V4-Flash-Vision-Exp; test a tool-result round trip before calling it coding-ready. Report any unsupported or untested models. 7. Tell me to restart Kimi, use /model, and select DeepSeek-V4-Flash-Vision-Exp (ChineseModels). Also give me the direct launch command: kimi -m "chinesemodels/DeepSeek-V4-Flash-Vision-Exp". 8. For a model-list refresh, rerun the installer; it updates only its managed block and retains my other models. Do not change my default without my request.
2. Or install from PowerShell
- Download the ZIP above, extract it, and inspect the included scripts. Open PowerShell in that folder.
- Run the command below and enter your ChineseModels API key at the hidden prompt.
- The installer backs up config.toml and updates only its ChineseModels block. Existing models, logins and your default remain in place.
powershell -NoProfile -ExecutionPolicy Bypass -File .\install-kimi-windows.ps1
kimi doctor config kimi -m "chinesemodels/DeepSeek-V4-Flash-Vision-Exp"
Kimi uses ~/.kimi-code/config.toml, or the directory set by KIMI_CODE_HOME. Its native api_key field stores the key in that local file as plain text. To manage the credential through your shell instead, use api_key_env and remove api_key; do not set both.
Download SHA-256 checksum. The archive contains source and installer files, with no customer credentials.
3. Select a model and thinking mode
Close and reopen Kimi Code, start a fresh conversation, then type /model in the terminal. Look for the “(ChineseModels)” entries. Their command-line aliases begin with chinesemodels/. Model IDs are case-sensitive; use the exact ID from your authenticated catalog.
Kimi forwards standard reasoning_effort values in native OpenAI-compatible mode. In our local wire check, DeepSeek, Qwen, MiniMax and Kimi requests carried the chosen effort, but did not carry vendor-specific thinking.type or enable_thinking fields.
Thinking limitation: a visible thinking toggle does not prove that a model's upstream reasoning can be disabled. Always-thinking models remain marked accordingly. For toggle-only families, “high” is a picker label, not a verified depth setting. Validate the behavior you need; this native installer does not normalize vendor-specific parameters.
4. Refresh, test and troubleshoot
Rerun the Kimi installer from the extracted folder to fetch the latest authenticated text-model list. It replaces only the block it manages, saves another backup, and leaves other providers intact. Run kimi doctor config, then restart Kimi.
- Authentication errors: check the key, account access and credit balance. Do not put the key in a support screenshot.
- Model not found: fetch /v1/models with your key and use the returned ID. Availability can change.
- Thinking or tool errors: use a supported option and test a text reply, then a tool call and its result. A text reply alone does not verify coding-agent compatibility.
- Restore: use the timestamped configuration backup. Review later edits before restoring so you do not discard unrelated changes.
Our October 22 check discovered 65 text/chat candidates and excluded 28 image/video generators. Your account may return a different list. Kimi's configuration validator passed and a live DeepSeek-V4-Flash reply succeeded. We have not certified every model's tool behavior in Kimi.
Scope and references
This guide targets the current Kimi Code client. Older kimi-cli installations use a different config location and provider schema. Do not paste these settings into a legacy installation without adapting them. Existing Kimi-managed search/fetch services remain separate from ChineseModels model access.
macOS and Linux users can use the prompt as a checklist, but the included PowerShell/DPAPI installer is Windows-only. A platform-appropriate credential store and launcher are required.