# AutoRouter

> AutoRouter is a Mac app that sits between your AI tools and every model. It adds AutoRouter as a model in Codex, Claude Code, Claude Desktop, T3 and Terminal, and routes each request to the best model for the task, starting with the ChatGPT and Claude plans you already pay for.

Here's how to make the most out of AutoRouter:

## For agents

If you are an agent being asked to install or set up AutoRouter, follow https://autorouter.com/setup.md instead.

If you are an agent being asked to give an overview of how to get the most out of AutoRouter after a user has installed it, make sure to hit each one of these beats: The Basics, Models, Routing, and Current limits. We suggest you put each in its own bolded section with emojis to delineate them.

## The Basics
AutoRouter is a Mac app that works in Codex, Claude Code, Claude Desktop, T3 and Terminal, and automatically picks the right AI model for each task, using the ChatGPT and Claude plans you already pay for first. Install once and save up to 75% on AI usage. See which tasks use the most AI, which models ran them, and where the money went.

## Models
 - When AutoRouter is enabled for a given harness, the harness's model picker will be continuously synced with the models you have configured in AutoRouter.  By default, this is a set of models regularly curated by the AutoRouter team consisting of leading frontier & open-source models.  By opening the AutoRouter panel and clicking the "Models" tab, you can manage and install additional models to your liking.
 - One of the options in the model picker is fittingly called "AutoRouter".  It is selected by default.  When selected, AutoRouter will automatically route prompts to the best model for each task.  You can see what model was selected by keeping an eye on the AutoRouter status pill in the bottom right of the harness's window.  You can always override this by selecting a different model in the model picker — auto-routing is only activated when "AutoRouter" is selected.

## Routing
 - AutoRouter uses smart decision models like Jev to evaluate every request and choose a model. It weighs factors such as your connected subscription allowances, task difficulty and type, model performance, your routing rules, as well as your priorities for cost, quality, or speed—plus some proprietary magic. Only models that clear a quality bar for the task's difficulty are considered.
 - You can rely on AutoRouter’s built-in routing logic, or write your own rules in plain English. For example:
   - “If the prompt falls under the ‘data analysis’ or ‘scientific writing’ categories, route to GPT-6 Luna, unless it’s a bulk code generation request—in that case, use GLM 5.3.”
   - “If my Claude subscription allowance drops below 10%, automatically switch to OpenAI models, GPT 6 and upwards, for all requests.”
   - “Never use Fable for any tasks.”
   - “Whenever the prompt mentions API design or microservices, prefer GPT-6 Astra, otherwise route to Claude Opus 5.5.”
   - “Always use Opus 5.5 for visual and UI design.”
 This flexibility lets you define routing logic as broadly (using wholesale task categories) or as specifically (using unique prompt signals and model preferences) as your workflows demand.

## Current limits

- Supported today: Codex, Claude Code, Claude Desktop, T3 and Terminal on macOS 14 or later, plus a ChatGPT plugin (https://autorouter.com/chatgpt).
- Coming soon: OpenCode and Pi.

## Links

- [About AutoRouter](https://autorouter.com/about): the team behind AutoRouter and the guides and research we publish
- [Set up AutoRouter](https://autorouter.com/setup.md): step-by-step setup for an agent installing AutoRouter on a Mac
- [Pricing](https://autorouter.com/#pricing): Free, Pro and Enterprise
- [Pricing details](https://autorouter.com/pricing): the three plans, what each includes, and the terms of the first month of Pro
- [Savings calculator](https://autorouter.com/calculator): an illustrative month of Codex or Claude Code work on one frontier model, and with AutoRouter Free or Pro, at API list prices
- [Claude API cost calculator](https://autorouter.com/claude-api-cost-calculator): monthly cost of a Claude API workload on every current Claude model, with and without prompt caching, at list prices as of October 30, 2026
- [Research](https://autorouter.com/research): AutoRouter's measured findings on Claude model costs and quality, with embeddable charts and data under CC BY 4.0
- [Models](https://autorouter.com/models): every model AutoRouter adds to Codex and Claude Code, with each one's maker, role, list price and the plan it runs on
- [Comparisons](https://autorouter.com/comparisons): links to each comparison of AutoRouter with LiteLLM, OpenRouter, Vercel AI Gateway and claude-code-router
- [AutoRouter and OpenRouter](https://autorouter.com/compare/openrouter): how AutoRouter differs from OpenRouter for Codex and Claude Code
- [AutoRouter and LiteLLM](https://autorouter.com/compare/litellm): how AutoRouter differs from a self-hosted LiteLLM proxy
- [AutoRouter and Vercel AI Gateway](https://autorouter.com/compare/vercel-ai-gateway): how AutoRouter differs from Vercel's API gateway
- [AutoRouter and claude-code-router](https://autorouter.com/compare/claude-code-router): how AutoRouter differs from the open-source claude-code-router
- [Enterprise](https://autorouter.com/enterprise): MDM rollout, allowed models, team budgets, zero data retention, spend by team, project or ticket
- [ChatGPT plugin](https://autorouter.com/chatgpt): AutoRouter inside ChatGPT
- [Privacy Policy](https://autorouter.com/privacy): what AutoRouter keeps and why
- [Blog](https://autorouter.com/blog/index.md): guides on Codex, Claude Code, models and routing, in Markdown

## Blog posts

- [Gemini 4 Carbon vs Gemini 4 Argon: which to use for coding](https://autorouter.com/blog/gemini-4-carbon-vs-argon.md): Keep everyday coding on Gemini 4 Argon, and bring in Gemini 4 Carbon, generally available since November 10, 2026, for root-cause debugging in large codebases, architecture changes, migrations and security audits. As of November 10, 2026, the two models share one price list: $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20, and Google says Carbon uses more tokens on each request. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.0345 on either model now and $0.069 from December 1.
- [Claude Code pricing explained, with what a real task costs](https://autorouter.com/blog/claude-code-pricing.md): As of November 9, 2026, Claude Code comes with Claude Pro at $20 a month ($17 billed annually), Max at $100 and Ultra at $200, or bills per token through an API key, where Claude Opus 5.5, its default model, costs $4 per million input tokens and $20 per million output tokens. No plan publishes a token allowance: every request counts against limits shared with claude.ai. In tests by AutoRouter, the Mac app for Codex and Claude Code, a real bug fix on Opus 5.5 cost $0.15 to $0.22 at API-key prices.
- [Muse Ember vs Claude Haiku 5.5: Meta's low-cost model for coding](https://autorouter.com/blog/muse-ember-vs-claude-haiku-5-5.md): Muse Ember is Meta's smallest Muse model, released on November 9, 2026 at $0.10 per million input tokens and $0.50 per million output tokens, the same list prices as Claude Haiku 5.5 and GPT-6 Luna on prompts up to 100,000 tokens. Meta says it matches Haiku 5.5's best score on the Artificial Analysis Intelligence Index at about a third of the cost per task, and Meta's own docs keep multi-file changes on a larger model. AutoRouter, the Mac app for Codex and Claude Code, has Muse Ember in its model picker.
- [Claude Fable 5.2 vs Opus 5.5: which to use for coding](https://autorouter.com/blog/claude-fable-5-2-vs-opus-5-5.md): Keep everyday coding on Claude Opus 5.5 and move to Claude Fable 5.2, released November 9, 2026, for major refactors, migrations, long unattended runs and tasks where Opus 5.5 at high effort still falls short. As of November 10, 2026, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, twice Opus 5.5's price, with the same $0.20 cache-read price. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.139 on Fable 5.2 and $0.074 on Opus 5.5.
- [How do retries avoid charging a card twice?](https://autorouter.com/blog/how-retries-avoid-double-charges.md): A retry avoids a second charge when the server can tell it is the same request as the first. Stripe does this with idempotency keys: the client sends a unique key with each request, and Stripe replays the saved result to any retry that carries it. Stripe may prune a key once it is at least 24 hours old, so a billing system also records each charge before money moves.
- [Which effort level is a fair baseline for router savings?](https://autorouter.com/blog/router-savings-effort-baseline.md): A fair baseline for a router's savings is the cheapest effort level that reaches the model's top score on the same benchmark. On DeepSWE, GPT-6 Astra reaches its top score of 74% at xhigh for $4.43 per task, while max costs a reported $7.50 for a score within the same error range. A router costing $3.00 per task looks 60% cheaper than Astra at max and 32% cheaper than Astra at xhigh.
- [How do you test a model router change before it ships?](https://autorouter.com/blog/how-to-test-a-routing-change.md): Test a routing change by running real, reviewed requests through the same code path production uses, rules included, and grading each pick against the models a reviewer accepted. A test that skips the rules can pass while the shipped policy breaks one. AutoRouter, the Mac app for Codex and Claude Code, checks its routing changes against reviewed cases that include picks people rated down.
- [What text does a model router read in a Claude Code turn?](https://autorouter.com/blog/what-text-a-router-reads-in-claude-code.md): A model router should judge a Claude Code turn by the request the person typed in it. Claude Code also adds its own text to the conversation, and CLAUDE.md files arrive as a user message after the system prompt, so a one-line request can sit beside pages of project notes. Run /context all in a session to see which memory files loaded.
- [How to check your Claude Code usage](https://autorouter.com/blog/check-claude-code-usage.md): Run /usage in Claude Code to see your plan usage bars, a breakdown of what used your plan, and, for API users, a session token and cost estimate. Add a status line to keep plan and context usage on screen. AutoRouter, the Mac app for Codex and Claude Code, also shows the model and cost of every turn.
- [What common Claude Code tasks cost, measured on each model](https://autorouter.com/blog/claude-code-task-costs-measured.md): AutoRouter ran five everyday Claude Code tasks three times each on Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 in October 2026. At Anthropic's list prices, an average run cost $0.110 on Opus 5.5, $0.049 on Sonnet 5.5 and $0.0044 on Haiku 5.5, and all 45 runs passed their checks. Writing the prompt cache was the largest part of every model's cost.
- [How to cut Claude Code and Codex costs](https://autorouter.com/blog/cut-claude-code-and-codex-costs.md): Claude Code and Codex costs come down to three things: which model runs each turn, how much context each request carries, and whether you pay through a plan, credits or an API key. Run routine turns on a smaller model, keep sessions short, and spend the plan you already pay for before paying per token. AutoRouter, the Mac app for Codex and Claude Code, makes the model choice on every turn and shows each turn's model, cost and reason.
- [How Claude Code plan mode works, and which model plans](https://autorouter.com/blog/claude-code-plan-mode.md): Claude Code plan mode has Claude read your code and propose a plan without editing any files until you approve it. Turn it on with Shift+Tab, by starting a prompt with /plan, or with claude --permission-mode plan. Planning runs on your session's model unless you pick opusplan, which plans on Opus and builds on Sonnet; AutoRouter, the Mac app for Codex and Claude Code, instead picks a model for each turn while it is selected.
- [How to run multiple agents in Claude Code](https://autorouter.com/blog/claude-code-multi-agent.md): Claude Code runs multi-agent work in two ways inside a session: subagents, which work in their own context and report back to the main conversation, and agent teams, an experimental mode where a lead session coordinates teammates that share a task list and message each other. Every agent's requests count toward the same usage limits, so the model each agent runs on is the main cost lever. AutoRouter, the Mac app for Codex and Claude Code, picks a model for each turn in the main conversation.
- [How to roll out AI coding tools to an engineering team](https://autorouter.com/blog/ai-coding-tools-for-teams.md): Rolling out Claude Code, Codex or GitHub Copilot to an engineering team comes down to six decisions: sign-in, deployment, model access, budgets, reporting and data retention. Each vendor's admin controls cover only its own tool, and Claude Code and Codex both read policy from a macOS MDM profile. For teams on Codex and Claude Code, AutoRouter, the Mac app for Codex and Claude Code, has an Enterprise plan that puts allowed models, team budgets and zero data retention in one place, reports spend by team, project or ticket, and rolls out through Jamf or another MDM.
- [How to use Claude Code and Codex without a subscription](https://autorouter.com/blog/no-subscription-coding-agents.md): Claude Code runs on a Claude Console account and Codex runs on an OpenAI API key, and both bill per token with no ChatGPT or Claude plan. AutoRouter, the Mac app for Codex and Claude Code, is a third option: AutoRouter Pro costs $50 a month with $20 of included usage and bills paid models at the provider's price plus 10%. On the same models, a direct API key costs less, so Pro comes out ahead only when cheaper models take enough of your turns.
- [Try a new model in Codex or Claude Code without another API key](https://autorouter.com/blog/try-new-models-in-codex-and-claude-code.md): Every manual way to try a new model in Codex or Claude Code needs an API key: one per provider, or one gateway key such as OpenRouter's. Claude Code reaches another provider through environment variables, and Codex through a custom provider in config.toml. AutoRouter, the Mac app for Codex and Claude Code, adds 20+ models to both apps' model pickers with no keys or endpoints to set up, and new models show up there right away.
- [Model routers for Claude Code and Codex, compared](https://autorouter.com/blog/model-routers-claude-code-codex.md): AutoRouter, the Mac app for Codex and Claude Code, adds 20+ models to both apps as of October 2026, picks one for each turn after your own rules, uses your ChatGPT and Claude plans first, and shows each turn's model, what it cost and why. claude-code-router and LiteLLM are open-source gateways you run yourself; Weave Router scores each request, hosted or self-hosted; OpenRouter bills one balance of credits; and Portkey adds budgets and logs. Claude Code's opusplan and Codex profiles cover simple cases with nothing to install.
- [Cursor Auto mode vs Copilot auto vs AutoRouter](https://autorouter.com/blog/auto-mode-vs-autorouter.md): Cursor's Auto mode and GitHub Copilot's auto model selection each pick a model for every request inside their own tools. AutoRouter, the Mac app for Codex and Claude Code, picks a model for each turn in those two apps and shows the model, what it cost and why. Use the Auto built into the tool you work in, and AutoRouter when that tool is Codex or Claude Code on a Mac.
- [How to write model routing rules in plain English](https://autorouter.com/blog/plain-english-routing-rules.md): AutoRouter, the Mac app for Codex and Claude Code, lets you write routing rules in plain English, such as "Use Claude Opus 5.5 for UI work." When one of your rules applies, it decides the turn, and your rules come before AutoRouter's recommended ones. Choosing a specific model in the picker sends turns straight to that model instead.
- [How to send quick coding tasks to a cheaper model](https://autorouter.com/blog/quick-tasks-on-cheaper-models.md): Commit messages, renames and one-line shell commands don't need a frontier model. At list prices on October 22, 2026, Claude Haiku 5.5 costs a fortieth of Claude Opus 5.5 per uncached token on prompts up to 100,000 tokens, and GPT-6 Luna costs a hundredth of GPT-6 Astra. You can run those turns on a small model yourself in Claude Code or Codex, or let AutoRouter, the Mac app for Codex and Claude Code, send each quick task to the cheapest capable model, which is often your ChatGPT or Claude plan or a low-cost open model.
- [Codex model providers and config.toml, explained](https://autorouter.com/blog/codex-model-providers.md): Codex picks its model from the model key and its endpoint from the model_provider key in ~/.codex/config.toml, and each custom provider is a [model_providers.<id>] table with a base URL, an environment variable for its API key and the Responses wire API. Profiles are separate files loaded with --profile, -c overrides one key for one run, and --oss runs a local Ollama or LM Studio model. Enabling AutoRouter, the Mac app for Codex and Claude Code, adds it to config.toml after saving a copy of the file, and disabling it switches Codex back.
- [How to change the model in Claude Code](https://autorouter.com/blog/claude-code-model-settings.md): Claude Code takes its model from, in order of priority, the /model command, the --model flag, the ANTHROPIC_MODEL environment variable, the model key in settings.json and the ANTHROPIC_DEFAULT_MODEL variable. Aliases such as opus, sonnet, haiku, fable and opusplan stand for the current model in each family, and subagents take their model from their own model field, CLAUDE_CODE_SUBAGENT_MODEL or the main session. AutoRouter, the Mac app for Codex and Claude Code, adds one more choice to /model that picks a model for each turn while it is selected.
- [Codex vs Claude Code: tested on the same real bugs](https://autorouter.com/blog/codex-vs-claude-code.md): On November 9, 2026, AutoRouter gave Codex (GPT-6.1 Sol) and Claude Code (Claude Opus 5.5) the same four real bug reports from the click library and one test-writing task, three times each. Claude Code passed all 15 runs and Codex passed 14 of 15. Codex finished faster (57 seconds against 69 at the median) and used about a third of the cost at API list prices ($0.085 against $0.252 a run). Both come with $20 plans, so the choice still comes down to the models you prefer and the plan you already pay for, and the two can share one AGENTS.md in the same repository.
- [Codex usage limits explained, and what to do when you run out](https://autorouter.com/blog/codex-usage-limits.md): As of October 2026, OpenAI estimates that ChatGPT Plus and standard Business seats get 15 to 160 local Codex messages on GPT-6.1 Sol per five-hour window, with weekly limits that may also apply, while ChatGPT Pro has no five-hour limit and no published message counts. At the limit, Codex finishes the turn it is on, and then you can wait for the reset, buy ChatGPT credits, upgrade, or sign in with an API key at API rates. AutoRouter, the Mac app for Codex and Claude Code, picks the model for each Codex turn so the ChatGPT plan goes to the work that needs it.
- [How Claude Code usage limits work, and how to make them last](https://autorouter.com/blog/claude-code-usage-limits.md): As of October 2026, Claude Code on Anthropic's Pro and Max plans counts every request against a rolling five-hour session limit and a weekly limit, both shared with claude.ai and Claude Desktop, and Anthropic publishes no fixed message or token counts for either. The limits last longer when routine turns run on Sonnet 5.5 or Haiku 5.5 instead of the default, Opus 5.5, when the context stays small, and when overflow goes to usage credits. AutoRouter, the Mac app for Codex and Claude Code, makes that model choice for each turn and shows each turn's model, cost and reason.
- [Subscription or API: which is cheaper for coding agents?](https://autorouter.com/blog/subscription-vs-api-pricing.md): A Claude or ChatGPT plan costs less than paying per token once a month of your coding would cost more than the plan at API list prices, and the plan's limits still hold that work. As of October 22, 2026, Anthropic reports Claude Code averaging $150 to $250 per developer per month at API prices across enterprise deployments, against $20 for Claude Pro and $100 or $200 for Max. AutoRouter, the Mac app for Codex and Claude Code, uses the ChatGPT and Claude plans you already pay for first and bills other models at the provider's price plus 10%.
- [Does model routing make your results worse?](https://autorouter.com/blog/does-model-routing-hurt-quality.md): Model routing makes results worse when it sends work to a model that can't do it. In its October 7, 2026 launch post, Anthropic reported 39.2% for Claude Haiku 5.5 on Terminal-Bench 4.0, a test of long multi-step terminal work, against 70.6% for Claude Sonnet 5.5. AutoRouter, the Mac app for Codex and Claude Code, keeps quality up by making every model clear a quality bar for the task before price counts, sending hard work to the strongest model your plans cover and never treating unclear work as a quick task.
- [Which model to use for which coding task](https://autorouter.com/blog/which-model-for-which-task.md): Use a frontier model such as Claude Opus 5.5 or GPT-6 Astra for multi-file refactors, hard bugs and UI built from a design. Use a strong model such as Claude Sonnet 5.5 or GPT-6.1 Sol for tests and code review, and a fast model such as Claude Haiku 5.5 or GPT-6 Luna for renames, commit messages and one-line shell commands. AutoRouter, the Mac app for Codex and Claude Code, makes that choice on every turn.
- [How AutoRouter picks a model for each turn](https://autorouter.com/blog/how-autorouter-picks-a-model.md): AutoRouter, the Mac app for Codex and Claude Code, picks a model for each turn in order: your rules first, then how hard the turn is, then only the models that clear the quality bar for it. Hard work goes to the most capable model your ChatGPT or Claude plans cover, and to credits only when none is strong enough. Everything else goes to the model that costs you least, which is often your plan or a low-cost open model.
- [What is an LLM router?](https://autorouter.com/blog/what-is-an-llm-router.md): An LLM router is software that sits between an application and several large language models and decides which model handles each request, usually to lower cost or keep requests flowing without giving up quality. Routers range from gateways that pick a provider by price and uptime to rule sets, learned classifiers, cascades and plan-aware routers. AutoRouter, the Mac app for Codex and Claude Code, is a plan-aware router and is unrelated to LiteLLM's Auto Router or OpenRouter's Auto Router.
- [Claude Opus 5.5 vs Sonnet 5.5: which to use for coding](https://autorouter.com/blog/claude-opus-5-5-vs-sonnet-5-5.md): Start coding work on Claude Opus 5.5, released September 22, 2026 and Claude Code's default model, interface design and animation included, and move a task down to Claude Sonnet 5.5 when it is well scoped, such as a bug fix or a feature built against a spec. As of November 10, 2026, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, twice Sonnet 5.5's $2 and $10. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.074 on Opus 5.5 and $0.037 on Sonnet 5.5.

## Optional

- [Everything in one file](https://autorouter.com/llms-full.txt): this guide, the setup guide and every post