# Run Claude Code with a different model provider: the variables per vendor, what CC Switch and magpie write, and what breaks

> Source: <https://stackness.dev/blog/run-claude-code-with-a-different-model-provider-env-vars-what-breaks>
> Published: 2026-10-14 08:51:48+00:00

# Run Claude Code with a different model provider: the variables per vendor, what CC Switch and magpie write, and what breaks

To run [Claude Code](https://stackness.dev/tools/claude-code) on another vendor's model, you set `ANTHROPIC_BASE_URL` to the vendor's Anthropic-compatible endpoint, put the vendor's key in `ANTHROPIC_AUTH_TOKEN` and map all four model tiers to the vendor's models, which takes about ten minutes. The steps below are copied from each vendor's docs on 14 October 2026, when Claude Code 2.1.299 was the latest release; some vendors last tested older versions. Anthropic's [gateway docs](https://code.claude.com/docs/en/llm-gateway) say it does not support routing Claude Code to non-Claude models, so when something breaks, the vendor's docs and issue tracker are where the fix is.

The [harness lock-in post](https://stackness.dev/blog/deepseek-harness-vs-claude-code-which-harnesses-lock-you-to-one-model-and-what-breaks-when-you-switc) covers what you leave behind in plugins, hooks and sessions. This one is the setup.

## What you need

- Claude Code 2.1.268 or later. Versions 2.1.265 to 2.1.267 sent a tool schema that Z.ai and DeepSeek rejected on every interactive request ([issue #92964](https://github.com/anthropics/claude-code/issues/92964) ).
- A key for the vendor. At Kimi, pay-as-you-go keys and the coding plan are separate products with different endpoints, and MiniMax's endpoint below takes an M Plan key.
- For a local model, [Ollama](https://stackness.dev/tools/ollama) , with its context length set to 64k or more.

| Vendor | Endpoint for `ANTHROPIC_BASE_URL` | Model the vendor's docs put on the top tiers | 
|---|---|---|
| DeepSeek API | `https://api.deepseek.com/anthropic` | `deepseek-flash[1m]` | 
| Kimi API | `https://api.moonshot.ai/anthropic` | `kimi-k3[1m]` | 
| Kimi Code plan | `https://api.kimi.ai/coding/` | `k3-256k` or`k3[1m]` | 
| Z.ai GLM Coding Plan | `https://api.z.ai/api/anthropic` | `glm-5.3[1m]` | 
| MiniMax M Plan | `https://api.minimax.io/anthropic` | `MiniMax-M3.1-Flash-Preview[1m]` | 
| [OpenRouter](https://stackness.dev/tools/openrouter) | `https://openrouter.ai/api` | `~anthropic/claude-sonnet-latest[1m]` , or any model with`CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1` | 
| Ollama | `http://localhost:11434` | Any local model | 

Sources: [DeepSeek](https://api-docs.deepseek.com/quick_start/agent_integrations/claude_code), [Kimi API](https://platform.kimi.ai/docs/guide/claude-code-kimi.md), [Kimi Code](https://www.kimi.com/code/docs/en/third-party-tools/claude-code.html), [Z.ai](https://docs.z.ai/devpack/tool/claude), [MiniMax](https://platform.minimax.io/docs/m-plan/claude-code.md), [OpenRouter](https://openrouter.ai/docs/guides/guides/claude-code-integration), [Ollama](https://docs.ollama.com/integrations/claude-code).

## 1. Test the endpoint with curl first

Testing the endpoint before Claude Code starts separates a wrong URL or key from a Claude Code problem. Anthropic's [connection guide](https://code.claude.com/docs/en/llm-gateway-connect) gives this request:

```
curl -sS -w '\n%{http_code}\n' -X POST "$ANTHROPIC_BASE_URL/v1/messages" \
  -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "max_tokens": 1, "messages": [{"role": "user", "content": "."}]}'
```

An error naming an unknown model still means the URL and key work. A `401` means the key was rejected. The base URL ends before `/v1`: adding `/v1` yourself produces `/v1/v1/messages`.

## 2. Set all four model tiers in settings.json

Claude Code calls four model tiers, not one: Opus, Sonnet and Fable for the main work, Haiku for background calls, plus a separate model for subagents. Kimi's docs warn that setting only some of them makes the rest fail silently. DeepSeek's snippet does not set the Fable tier (`ANTHROPIC_DEFAULT_FABLE_MODEL`), and DeepSeek maps any model name it does not support to `deepseek-flash`. DeepSeek's setup, moved into the `env` block of `~/.claude/settings.json`:

```
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "<your DeepSeek API key>",
    "ANTHROPIC_MODEL": "deepseek-flash[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-flash[1m]",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-flash[1m]",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-flash",
    "CLAUDE_CODE_SUBAGENT_MODEL": "deepseek-flash",
    "CLAUDE_CODE_EFFORT_LEVEL": "max",
    "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "786432"
  }
}
```

Use `~/.claude/settings.json` for every project or `.claude/settings.local.json` for one, never the committed `.claude/settings.json`. When a shell export and the settings file set the same variable, the settings file wins, so a leftover `env` block overrides whatever you export later. Use `ANTHROPIC_AUTH_TOKEN` unless the vendor says otherwise; Kimi's coding plan is the exception and uses `ANTHROPIC_API_KEY`. `ANTHROPIC_SMALL_FAST_MODEL`, which older guides set, is [deprecated](https://code.claude.com/docs/en/model-config) in favour of `ANTHROPIC_DEFAULT_HAIKU_MODEL`.

For Ollama, `ollama launch claude` does all of this for one run: it sets the variables on the child process only and writes no Claude Code settings file ([source](https://github.com/ollama/ollama/blob/main/cmd/launch/claude.go)).

## 3. Check /status, then the vendor's dashboard

`/status` in Claude Code confirms the switch: it shows an `Anthropic base URL` line only when a gateway is set, and an `Auth token: ANTHROPIC_AUTH_TOKEN` line when the key is the one in use. A `Login method` line naming claude.ai means your vendor key is not active. Then run `/model` and check that each tier lists the vendor's model.

Read the bill on the vendor's dashboard, not the dollar figure in `/usage` (`/cost` is now an alias for it), which Claude Code computes locally from token counts at list price ([costs](https://code.claude.com/docs/en/costs)). The model names mislead too: DeepSeek maps `claude-opus` requests to `deepseek-v4-pro` and bills the V4 Pro price ([DeepSeek](https://api-docs.deepseek.com/guides/anthropic_api)), and since about 9 October its endpoint echoes the requested model back, so transcripts record Opus ([issue #101637](https://github.com/anthropics/claude-code/issues/101637)). `/usage` does show the prompt cache hit ratio, which is the number to watch.

## What CC Switch and magpie write to disk

A switcher does steps 2 and 3 for you, and it rewrites your config each time you switch. [CC Switch](https://stackness.dev/tools/cc-switch) v4.0.11, released 13 October with 144,049 GitHub stars, and [magpie](https://stackness.dev/tools/magpie) v0.1.1233, released 14 October with 12,231 stars, are both MIT-licensed and touch the same file.

|  | CC Switch | magpie | 
|---|---|---|
| Claude Code file it edits | `env` keys and`model` in`~/.claude/settings.json` ; leaves permissions, hooks and plugins alone | `env` keys in`~/.claude/settings.json` ; restores the old values when you switch back | 
| [OpenAI Codex](https://stackness.dev/tools/codex-cli) | Writes `[model_providers.custom]` to`~/.codex/config.toml` and deletes`~/.codex/auth.json` on a direct switch, stashing the login | Adds `[model_providers.magpie]` and a model catalog file | 
| Local proxy | `127.0.0.1:15721` in Routing mode, which writes the placeholder key`PROXY_MANAGED` | Gateway at `127.0.0.1:3425` , translating between API formats | 
| Its own data | `~/.cc-switch/` (SQLite database, backups every 24 hours) | `~/.config/magpie/` (keys in`providers.json` , replaced values in`stash.json` ) | 

Sources: [CC Switch config file docs](https://github.com/farion1231/cc-switch/blob/main/docs/user-manual/en/5-faq/5.1-config-files.md), [magpie reference](https://github.com/yetone/magpie/blob/main/docs/reference.md), [magpie README](https://github.com/yetone/magpie).

Three side effects to know. CC Switch does not save the model you picked with `/model` back to the previous provider. magpie's docs say a key in `ANTHROPIC_AUTH_TOKEN` signs Claude Code out of claude.ai while it runs through magpie, which turns off claude.ai's plan limits in `/usage`, its connectors, voice and `/teleport`, and that Remote Control stays off because Claude Code offers it only on Anthropic's own address. magpie also reads your existing Claude Code and Codex sign-ins to offer them as providers (`magpie accounts discovery off` stops it) and sends one usage ping a day unless `DO_NOT_TRACK=1` is set.

## What goes wrong, and the fix

### The cache misses and the bill jumps

Prompt caching is where a provider swap costs or saves the most, because a long session resends the whole context every turn. Anthropic's [caching docs](https://code.claude.com/docs/en/prompt-caching) warn that a gateway which drops the cache markers "bills as uncached input on every turn". DeepSeek ignores the markers and caches on its own, best-effort, so hits are not guaranteed. OpenRouter keeps a session on one provider for its cache until 10 minutes of inactivity. Ollama does not support the markers, which costs nothing on a local model.

Input cost of one turn that resends a 100,000-token context, at each vendor's list price on 14 October 2026 (output and cache writes left out):

| Model | Uncached | Cached | 
|---|---|---|
| [Claude Sonnet](https://stackness.dev/tools/claude-sonnet) 5.5 | $0.20 | $0.01 | 
| [DeepSeek](https://stackness.dev/tools/deepseek-v4) V4.1 Flash, peak hours | $0.03 | $0.0006 | 
| DeepSeek V4 Pro, peak hours | $0.132 | $0.0044 | 
| [Kimi K3](https://stackness.dev/tools/kimi-k3) | $0.30 | $0.03 | 
| [GLM](https://stackness.dev/tools/glm) -5.3, pay-as-you-go | $0.14 | $0.026 | 
| [MiniMax-M3](https://stackness.dev/tools/minimax-m3) | $0.03 | $0.006 | 

Prices: [Anthropic](https://platform.claude.com/docs/en/about-claude/pricing), [DeepSeek](https://api-docs.deepseek.com/quick_start/pricing), [Kimi](https://platform.kimi.ai/docs/pricing/chat), [Z.ai](https://docs.z.ai/guides/overview/pricing), [MiniMax](https://platform.minimax.io/docs/pricing/overview). DeepSeek halves both prices off-peak. A missed cache on DeepSeek V4 Pro ($0.132) or even V4.1 Flash ($0.03) costs more than a hit on Sonnet 5.5 ($0.01), so a cheaper model with a broken cache can cost more than the Claude model it replaced. Kimi K3 lists input above Sonnet 5.5, at $3 against $2 per million tokens.

**Fix:** watch the hit ratio in `/usage` for the first sessions, and do not switch provider in the middle of a session: CC Switch's docs note that the new model has to rebuild its prompt cache.

### Background calls fail with "invalid thinking"

Some models only run with thinking on. Kimi's docs show `kimi-k2.7-code`, which their own setup puts on the Haiku tier, rejecting requests sent with thinking off with `400 invalid thinking: only type=enabled is allowed for this model`, and tell you to keep thinking on in Claude Code. MiniMax's M3.1 Flash Preview also cannot turn thinking off. CC Switch hit this when its Routing mode wrote the Haiku tier as `claude-haiku-4-5`, a name Claude Code treats as able to run without thinking, and [fixed](https://github.com/farion1231/cc-switch/issues/8118) it on 12 October.

**Fix:** keep thinking on, or put a model that accepts thinking off on `ANTHROPIC_DEFAULT_HAIKU_MODEL`, and update the switcher.

### /context shows a 200k window on a 1M model

Claude Code assumes 200,000 tokens for a model name it does not know, so it compacts early or reports a nearly full context before the first message ([issue #94487](https://github.com/anthropics/claude-code/issues/94487)). Without a matching auto-compact window, it can also run into the vendor's real limit.

**Fix:** use the `[1m]` suffix the vendor's docs give, or set `CLAUDE_CODE_MAX_CONTEXT_TOKENS` and `CLAUDE_CODE_AUTO_COMPACT_WINDOW` to the model's real window ([env vars](https://code.claude.com/docs/en/env-vars)). An ID with `[1m]` ignores `CLAUDE_CODE_MAX_CONTEXT_TOKENS` unless `CLAUDE_CODE_DISABLE_1M_CONTEXT=1` is also set.

### Requests fail after a Claude Code update

New Claude Code releases send new tool schemas and beta fields that other vendors' endpoints may reject. Anthropic's troubleshooting table gives `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` for an `Extra inputs are not permitted` error. For a schema error that starts with one release, check the vendor's and the switcher's issue trackers before downgrading.

### Images and web search stop working

DeepSeek V4 Pro does not take images; only Flash does. Claude Code's WebSearch tool runs on Anthropic's search backend, which DeepSeek's endpoint serves itself at extra token cost; Z.ai and MiniMax offer search through MCP servers instead.

## When the vendor's own harness is the better choice

If you will run one vendor's model full time, its own harness avoids most of these failures: [ZCode](https://stackness.dev/tools/zcode) for GLM, [DeepSeek Harness](https://stackness.dev/tools/deepseek-harness) (a preview), [Kimi Code](https://stackness.dev/tools/kimi-code) for Kimi, or a model-agnostic one such as [Pi](https://stackness.dev/tools/pi), all among the [AI coding tools developers list](https://stackness.dev/categories/ai-tools). The swap makes sense when you want Claude Code's hooks, skills and plugins and a cheaper model for part of the work, which is the trade the [model routing post](https://stackness.dev/blog/model-routing-for-coding-agents-the-plan-and-price-numbers-developers-measured-in-september-2026) prices. On Stackness, as of 14 October 2026, 8 real profiles list Claude Code, 1 lists Kimi K3 and none list CC Switch or magpie, after removing bot accounts ([data sources](https://stackness.dev/about/data-sources)), so these are early setups.

## Checklist

```
[ ] Claude Code 2.1.268 or later
[ ] curl test returns 200 or an unknown-model error, not 401
[ ] ANTHROPIC_BASE_URL without /v1, key in ANTHROPIC_AUTH_TOKEN (Kimi Code: ANTHROPIC_API_KEY)
[ ] Opus, Sonnet, Fable, Haiku and subagent models all set, in ~/.claude/settings.json or settings.local.json
[ ] Haiku-tier model works with the thinking setting you run
[ ] [1m] suffix or CLAUDE_CODE_MAX_CONTEXT_TOKENS matches the real window
[ ] /status shows the base URL and the auth token line
[ ] cache hit ratio checked in /usage; bill read on the vendor's dashboard
[ ] no provider switch mid-session
```

## Which of these do you use?

Nobody on Stackness runs OpenRouter yet. Tick what you use and be the first.

[Play Lonny Eagle Amateur Skater 4A pixel art skateboarding game. Signing up opens it straight away.Sign up to play](https://stackness.dev/games/lonny-eagle)