Qwen 3.8
Free
Your subscriptions only
$0/mo
Free to route across your
subscriptions
- Supported harnesses
- Up to 2 subscriptions
- Personal routing rules
- Models outside your plans
Every lab’s best model,
in any harnessCut your AI bill in half
across every harness
Every frontier model.
Half the engineering
AI spend
Govern every AI harness.
Cut your token bill
in half
One routing layer across
every
engineer’s Mac
Stop burning frontier quota on routine turns or hitting usage walls mid-afternoon. AutoRouter pools your Claude and ChatGPT plans, brings every leading model into the coding tools you already use, and routes each turn to the model built for it—so your plans go twice as far.Deploy intelligent model routing to every engineer’s Mac in one MDM push. Keep complex architecture and debugging on frontier models, move routine turns to fast models that clear the same bar, and govern every harness with team budgets, spend attribution, and strict zero data retention. Give every developer the strongest model from every lab inside Codex, Claude Code, and Terminal—without API key sprawl or runaway token bills. AutoRouter routes each turn by difficulty and policy, enforces zero data retention, and cuts wasted frontier spend across the org. Your engineers shouldn’t burn frontier tokens on file reads and renames—or get throttled mid-sprint. Roll out AutoRouter in one MDM push to govern every coding harness, route every turn to the model that fits, and attribute every dollar by team and project.
Download for macOS
Free · macOS 14 or later
Contact sales
One-push MDM · Zero data retention
Claude Fable 5.1GPT-6 AstraGPT-6.1 SolClaude Opus 5.5Claude Sonnet 5.5Claude Haiku 5.5GPT-6 SolGPT-6 Luna
Gemini 4 Argon
Grok 4.7
DeepSeek V4.1 Flash 1017
GLM 5.3
Kimi K3
MiniMax M3.1 FlashGLM 5.3 Flash
Same performance.
Half the costHalf the AI spend.
Frontier quality where it counts
Frontier models for hard code.
Fast models for
routine turns
Cut wasted frontier spend
without slowing
engineers down
Sticking to a single model means overpaying on easy turns and settling whenever another lab is stronger. AutoRouter matches every turn in your session to the model that handles it best, saving your frontier plans for the work that needs them.Over half of an engineering session is file reads, boilerplate, renames, and commit messages that never needed your priciest model. AutoRouter evaluates every prompt in flight, locks complex architecture and debugging onto frontier models, and routes routine turns to fast models only when they clear the same quality bar. Standardizing on one frontier model means paying top-tier prices for every file read, rename, and summary. AutoRouter classifies each turn in real time so complex engineering gets the strongest model across labs while routine turns run at a fraction of the cost.
Codex
Refactor the billing module to use the new ledger API
AutoRouter- On plan
-
Claude Opus 5.5Claude plan
-
GPT-6 AstraChatGPT plan
- Pay as you go
-
DeepSeek V4.1 FlashCredits
Examples. Real routes follow your plans and rules.
Install once. Every harness gets smarter
No new editor to learn and no API keys to wire up. AutoRouter drops straight into the model picker of the harnesses already on your Mac.
-
Download AutoRouter
Install the Mac app and sign in with the Claude and ChatGPT subscriptions you already pay for.
-
Set the model to AutoRouter
Select AutoRouter in Codex, Claude Code, Claude Desktop, T3, or Terminal. Recommended routing works out of the box.
-
Keep building
Every turn routes automatically in the background, with live model picks and usage tracked in your menu bar.
Make every subscription go twice as far
Even a $200 Max plan runs dry when every file read and
rename hits your top model. AutoRouter spreads work across
your
Claude
and
ChatGPT
subscriptions and moves mechanical turns to fast models, so
your usage limits last through Friday.
- More work on Free
- 1.9×
- More work on Pro
- 2.9×
Using
Full
Empty
Tokens
Frontier for everythingAutoRouter FreeAutoRouter Pro
Savings vary with your task mix and model choices. This
illustrative comparison shows how plans can last longer than
using a frontier model for every task; it is not a measured
customer result. The provider logos illustrate the range
available in each tier, not the providers used to calculate
these lines.
An example week of 30% hard, 45% medium and 25% quick turns,
at
Claude Opus, Sonnet and Haiku 5.5 list prices with cached input at a
tenth of input. Pro sends 85% of quick and 75% of medium turns
to open models, which its included usage pays for.
All your subscriptions, pooled into one
Paying for both a work and personal
Claude
or
ChatGPT
plan? Connect them all once in Settings. AutoRouter paces usage
across every account you own and rolls over seamlessly when any
single plan hits its limit.
Start Pro
Free includes up to 2 subscriptions
Settings
Plans
ChatGPT
2 plans connected
Claude
4 plans connected
Work
Personal
Team Max
Research
Account rotation modeEach new session goes to the account with the
most left.
Every pick explained.
Every token accounted forEvery turn traced.
Every dollar explained
Complete auditability across
every session
and seat
See exactly where your
AI budget goes—turn
by turn
Open any session in the Mac app to see the turn-by-turn breakdown: which model ran, why AutoRouter picked it, how long it took, and what it cost compared with keeping a single frontier model on every prompt.Give engineering leaders and developers a transparent audit trail of every session: which model ran each turn, why policy or difficulty selected it, and the exact savings against a single frontier model. No opaque routing decisions. Every session records the model, latency, token count, policy match, and cost delta for each turn—giving engineering and finance a verifiable record of both quality and savings.
Every leading lab. Zero API keys to manageEvery model provider, one governed catalog Every approved lab through one enterprise contract
Your harness shouldn’t lock you into a single lab. AutoRouter brings Anthropic, OpenAI, Google, xAI, DeepSeek, Kimi, Qwen, Z.ai, MiniMax, and more into every supported tool on your Mac.Give your teams frontier and open-weight models across every major lab through one enterprise account, with no per-provider procurement or API key sprawl. Approve the exact models and providers your teams can use across OpenAI, Anthropic, Google, and leading open-weight labs—all billed and governed in one place.
Rules in your own wordsOrg policy in plain English Centralized routing rules your whole org follows
Tell AutoRouter how you like to work—“Use Claude Opus for UI design” or “Never send customer code to open models.” Your rules override everything else, and you can pin a specific model on any turn.Start from our recommended routing defaults, then set org-wide or team-specific rules in plain English that engineering, security, and finance can audit at a glance. Define routing policies by repository, path, or team in plain English—such as locking payment code to frontier ZDR models—and enforce them automatically across every engineer’s Mac.
-
Use
GPT-6.1
Sol
for debugging.Yours
-
Never use
Kimi
K3
on customer code.Yours
-
Use
Claude Opus
5.5
for anything in /payments.Yours
-
Use
Claude Opus
5.5
for UI design.12 turns this week
-
Use
GPT-6
Astra
for debugging.Overridden by yours
-
Use
MiniMax M3.1
Flash
for quick edits.21 turns this week
-
Use
DeepSeek V4.1
Flash
for renames.38 turns this week
- Use your plans before credits.Every turn
- Keep the same model while a task continues.Every turn
Multi-agent orchestration
One intelligence layer across
every model and
every harness
Today your AI tools lose context the moment you switch windows, and no single model catches its own blind spots. Because AutoRouter sits across every harness and model lab on your Mac, we’re turning that routing layer into a shared intelligence layer across all your AI work.Today each coding tool operates in a silo, with no shared memory across harnesses and no independent check on high-risk code. Because AutoRouter sits across every harness and model provider on your Macs, we’re building the coordination and verification layer for how engineering orgs ship software. Relying on one model family to both author and review production code leaves every blind spot unchecked. Because AutoRouter sits across every harness and provider in your stack, we’re turning routing into multi-model verification and shared organizational context.
-
Adversarial multi-model review
Never let a single model grade its own homework. Have Claude write the implementation and GPT independently pressure-test the logic, security, and edge cases before you merge.
-
One brain across all your tools
Your preferences, repo conventions, and active task context travel with you across Codex, Claude Code, Claude Desktop, T3, and Terminal so every harness picks up right where you left off.
-
Coordinated multi-agent runs
Coordinate multiple models inside a single workflow—combining the strongest planner, the fastest parallel code editors, and the sharpest debugger on every feature.
Privacy & security
Privacy and security enforced at the routing layer Zero data retention across every harness and model
Every prompt across every coding harness passes through one policy boundary you control—so security, legal, and engineering never have to trade speed for compliance. Enforce zero data retention, approved model catalogs, and identity-based access across every coding tool on your Macs from a single control point.
-
Zero data retention, guaranteed
Lock routing to zero-retention endpoints where prompts, completions, and proprietary code are never stored, logged, or used to train models.
-
Approved models and harnesses only
Decide which providers, model weights, and coding harnesses each team can use. Unapproved models never appear in your engineers’ pickers.
-
SSO, SCIM, and no raw API keys
Provision seats through your identity provider and MDM without distributing shared provider API keys across developer laptops.
AutoRouter for Individuals
Trying AutoRouter on your own Mac first? Start on your own Mac in two minutes
Want to test AutoRouter on your own Mac before rolling it out to the org? Download the Mac app for free, pool your personal and work Claude and ChatGPT subscriptions, and see how every turn routes across Codex, Claude Code, and Terminal. Evaluate AutoRouter on your own workflow today. Routing across your personal and work Claude and ChatGPT subscriptions is free, and your first month of Pro includes $20 of model usage.
Free on the plans you already pay forBuilt to pay for itself in the first week Simple pricing for individuals and engineering orgs
Route across your
ChatGPT
and
Claude
subscriptions for free. Pro pools unlimited personal and work
accounts, unlocks every frontier and open-weight provider, and
includes $20 in monthly API credits—with your first month
free.Roll out Enterprise across your organization with one-push MDM
deployment, centralized model governance, zero data retention,
and spend attribution—or start on your own Mac with Free or Pro.
Priced per seat for engineering organizations, with one-push
MDM rollout, zero data retention, and full spend attribution
by team and project.
Pro
Every model and provider
$50/mo
$20 in included API credits each month*
- Every model via API credits
- Unlimited subscriptions
- Priority support from our team
- Early access to exclusive features
Plus everything in Free
Start ProEnterprise
Rolled out to your whole team
Custom
Priced for your team
- Deploy through Jamf or your MDM
- Org rules, models and budgets
- Spend by team, project or tag
- Zero data retention
- SSO, SCIM and audit log
Plus everything in Pro
Get in touch*Included API credits reset monthly and do not roll over. Connected subscriptions are always used first; paid models outside your plans are billed from included API credits at the provider’s list price plus 10%, and top-up credits never expire. New accounts start with a free month of Pro, including $20 in API credits.
Bring AutoRouter to your whole engineering orgEnterprise controls built for how engineers work Govern every harness from one control plane Built for the whole engineering org
Cut your company’s engineering AI bill in half while giving every developer the strongest model for every turn. Roll out to every Mac in one MDM push with shared routing policy, team budgets, clear spend attribution, and strict zero data retention.Give your teams the models they need across every harness, with one-push MDM rollout, shared routing rules, clear spending reports, and strict zero data retention. Equip every developer on every Mac with frontier and open-weight models through one enterprise account—complete with per-seat budgets, team spend attribution, and zero provider setup.
Every lab’s best model.
Twice the work from
your plansHalf the AI bill.
Every engineer on the right model
One MDM push.
Every harness governed
and routed
Download AutoRouter, connect the subscriptions you already pay for, and keep building in the harnesses you already use—with every turn routed to the model built for it.Roll out AutoRouter across your engineering org with one-push MDM deployment, full spend attribution, and strict zero data retention—while giving every engineer the best model for every turn. Talk with our team to deploy AutoRouter across your Macs, or download the app today to test routing on your own subscriptions first.
Free · macOS 14 or later
Questions
Setup, routing, billing and privacy. Anything else, email support@autorouter.com.
What is AutoRouter?
A Mac app that routes AI work across your harnesses. Choose AutoRouter and each task goes to the model that fits, using your connected subscriptions first, every model on Pro, and centralized MDM governance and zero data retention on Enterprise.
How does Zero Data Retention work on Enterprise?
When Zero Data Retention is enabled for your organization or a specific team, AutoRouter routes exclusively to provider endpoints with zero-retention guarantees where prompts, completions, and code are never logged, retained, or used for training. Any model that does not meet your policy is blocked automatically.
How do we deploy AutoRouter across our Macs?
Push the signed AutoRouter package and configuration profile through Jamf, Kandji, or any MDM. Engineers authenticate with SSO and immediately get routed models inside Codex, Claude Code, Claude Desktop, T3, and Terminal without configuring API keys locally.
Will AutoRouter make my AI worse?
No. Hard work always goes to a frontier model, and routine turns go to models that handle them just as well, so you get the same results as running the top model on everything for a fraction of the usage. Many tasks come out better, because each one goes to the model that’s strongest at it.
Which apps and Macs does it work with?
Codex, Claude Code, Terminal, T3, and Claude Desktop. OpenCode and Pi are coming soon. AutoRouter runs on macOS 14 or later.
How do I start?
Download AutoRouter, connect your subscriptions, and choose it in a supported harness. The app guides you through setup.
How does it choose a model?
It reads each turn for its difficulty and kind of work, then weighs your plans’ limits and your rules, then balances cost, quality and speed. Quality comes first: a turn only moves to a cheaper model when that model does it just as well. How AutoRouter picks a model explains the process; routing and quality and Auto mode vs AutoRouter cover the tradeoffs.
Can I still pick a model myself?
Yes. AutoRouter’s models appear in your app’s model picker beside the ones you already have. Choose any of them and your turns go straight to it; AutoRouter routes only while it’s the model selected.
How do routing rules work?
You start with AutoRouter’s recommended rules. Add your own the way you’d say them, like “Use GPT-6.1 Sol for debugging,” and where yours and ours differ, yours win. Rules are strong guidance rather than a guarantee; to always use one model, pick it in the model picker.
What does enabling change on my Mac?
AutoRouter updates the settings it needs for each supported harness, saving a copy first. Your other preferences stay in place. Disable a harness in AutoRouter’s settings to switch it back.
I already pay for Claude Max and ChatGPT Pro. Is it for me?
Yes, it’s built for you. Routing across your subscriptions is free, and Pro connects several Claude and ChatGPT accounts, personal and work included. Routine turns stop eating your frontier usage, and when one account hits its limit, new sessions continue on the next, so you get more from the plans you have before you buy another.
Does it use my ChatGPT and Claude subscriptions?
Yes, first. In Codex, OpenAI models run on your ChatGPT plan; in Claude Code, Claude models run on your Claude plan. Connect the other subscription in settings to use your Claude plan inside Codex, or your ChatGPT plan inside Claude Code. On Pro, add as many accounts as you have; when one hits its limit, new sessions move to the next.
Do I need a ChatGPT or Claude plan?
No. Pro gives you every model, with $20 of usage each month.
What happens when a plan runs out?
On Pro, your usage covers OpenAI and Anthropic models until the plan’s limit resets. With none left, those turns stop and ask you to top up, while turns your plans still cover keep working.
How much does it cost?
Routing across your ChatGPT and Claude subscriptions is free. Pro is $50 a month and includes $20 of monthly model usage, multiple connected Claude and ChatGPT accounts, and every supported provider. Paid models outside your plans are billed at the provider’s list price plus 10%, spent from your included usage first and then from non-expiring top-up credits. New accounts start with a free month of Pro, up to $20 of usage.
What does a Claude Code task cost?
In AutoRouter's October 28, 2026 study, an everyday Claude Code task cost an average of $0.110 on Claude Opus 5.5, $0.049 on Claude Sonnet 5.5 and $0.0044 on Claude Haiku 5.5 across 45 runs of five coding tasks at Anthropic's list prices. See the full results and the published method. These are model-level measurements, not a savings estimate for your workload.
How do credits work?
Pro includes $20 of usage each month. It resets monthly and doesn’t roll over. Top up any time: credits you buy never expire.
Does AutoRouter store my prompts and code?
Requests pass through AutoRouter’s servers on their way to each provider, and we don’t store transcripts. We keep the message that opens each turn, up to 4,000 characters, to review routing, along with records of the model chosen, tokens and cost. The Privacy Policy has the details.
How do team budgets and spend attribution work?
Set recurring budgets by team or seat on your own fiscal calendar, with instant exceptions when a launch needs extra capacity. Spend and routing savings break down by team, project, ticket, or custom tag so engineering and finance can explain every dollar.
What do teams get?
Everything in Pro, plus deployment through your MDM; org rules, models and budgets; spend reporting; zero data retention; and SSO, SCIM and an audit log. See AutoRouter for Enterprise or get in touch to roll it out.