Developers
Send Maestro your agents’ jobs.
Building an agent or an app? Every model through one door. Give Maestro a job. It breaks the job into tasks, routes each task to the right model and the right cloud, checks the work, follows your rules, and puts every task in the record. One balance for every model.
- GPT drafts, plans
- DeepSeek code
- Qwen extraction
- Fable specs
- Gemini images
- GLM translation
- Astra hard planning
- Llama cleanup
- Sonnet tests
- Opus review
- Mistral tagging
- Grok research
Your agent’s jobs, routed and checked
One door
One door. Every model.
300+ AI model endpoints across 14 leading model providers.
From OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, GLM, Kimi, MiniMax, Mistral, NVIDIA, Cohere, Perplexity and more. One key, one balance, one record.
- SpecsFable
- Drafting, planningGPT
- Code, summariesDeepSeek
- TestingSonnet
- ReviewOpus
- ResearchGrok
- ImagesGemini
- TranslationGLM
- ExtractionQwen
- CleanupLlama
- TaggingMistral
- Hard planningAstra
The board changes when a better model ships or a price moves. Your agent gets the change on the next task.
Three layers
Access. Routing. Execution.
- 3An AI execution model.
One job, every model it needs. Maestro breaks the job into tasks, routes each task, checks every result, retries only the failures, and hands back one finished job.
- 2Model routing.
One prompt, one chosen model. A router chooses the model for one call.
- 1Model access.
One key, many models. You choose the model for every call.
Not a model that does the work. A model that gets the work done.
What a task gets
Route. Check. Record.
Route.
Send a task. Maestro reads what kind of work it is and how hard, and routes it to the model that does it best at the lowest price, on the cloud that runs it best.
Check.
A test, a review or a rule that the result must pass, run by something that did not produce it. A failed check sends only that task to a stronger model. You pay the higher price for that one task, not the whole job.
Record.
Every task, with its model, the reason, the check and the cost. Sponsored tasks come off the bill.
Routers send a prompt. Maestro executes the job.
A router sends one prompt to one model and stops there. Maestro looks at the whole job: which model for each task, which cloud, whether the result passed, and what to do when it did not. Then it learns from the answer. Your agent decides what to do. Maestro executes it.
Your rules
Rules, written as a short list.
Set rules for one product or a whole workspace. Route work by cost, speed, quality and your own rules. Pin a model on any task.
Every rule, every field and every job file is in the docs. Routing rules
# Example. A routing rule for one product.
rule: northwick-store
prefer: sponsored # use sponsored models when they do the task well
data_location: [us, ca] # providers that keep data in the US or Canada
check:
code: tests + review # tests must pass, a second model reviews
copy: review
on_fail: next_model_up # re-route only the failed task
max_cost_per_task: auto # set by the balance
pin:
images: gemini # always use this model for imagesFive ideas
Built on five ideas.
- Task
- One piece of work with one result: a spec, a file of code, an image, a reply to a customer.
- Rule
- Where a task may go and how much it may spend. Set once, applied to every task it covers.
- Route
- Maestro’s choice of model for a task, made as the task starts.
- Check
- A test, a review or a rule that the result must pass, run by something that did not produce it.
- Record
- Every task with its model, the reason, the check and the cost.
One task
One task, followed through.
An abandoned-cart email for Northwick Candle Co., our example store, which sells in two currencies. Each step shows its model and its cost. One test fails. Only the fix moves to a stronger model.
| Step | Model | Why this model | Cost |
|---|---|---|---|
| Write the task spec | Fable | Best at specs | $0.22 |
| Plan the files | GPT | Drafting and planning | $0.03 |
| Write the code, four files | DeepSeek | Strong code at a low price | $0.04 |
| Write the tests | Sonnet | Testing | $0.05 |
| One test failed. Re-routed to the next model up. | Opus | Stronger model for the fix | $0.08 |
| Review the change | Opus | Review by a model that did not write the code | $0.07 |
| Write the email copy and subject lines | GPT | Copy | $0.01 |
| Merge and deploy a preview | ZB Cloud | Hosting | $0.00 |
| Total | $0.52 |
The same task with everything on one model: $1.45. List prices on 10 October 2026. The deploy step is part of hosting, charged separately; the 5% fee is separate too.
The prices
Same task. 100 times the price.
The best model for a task is rarely the priciest one. Maestro knows every one of these prices, and routes each task to the cheapest model that does it well.
| Model | Provider | Best at | Input $/1M | Output $/1M |
|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | Specs | 10.00 | 50.00 |
| GPT-6 Astra | OpenAI | Hard planning | 10.00 | 50.00 |
| Gemini Nano Banana 2.1 | Images | 1.50 | 30.00 per 1M image tokens | |
| Claude Opus 5.5 | Anthropic | Review | 4.00 | 20.00 |
| GPT-6.1 Sol | OpenAI | Drafting and planning | 2.00 | 10.00 |
| Claude Sonnet 5.5 | Anthropic | Tests | 2.00 | 10.00 |
| Grok 4.7 | xAI | Research | 2.00 | 6.00 |
| GLM-5.3 | Z.ai | Translation | 1.40 | 4.40 |
| DeepSeek-V4-Pro | DeepSeek | Summaries and code | 1.32 | 3.96 |
| Llama 4 Maverick | Meta, open weights | Cleanup | 0.19 | 0.65 |
| Mistral Small 4 | Mistral AI | Tagging | 0.15 | 0.60 |
| Qwen3.8-Flash | Alibaba Cloud | Extraction | 0.15 | 0.47 |
Published list prices per 1 million tokens, standard tier, short context, no caching or batch discount, retrieved 10 October 2026. DeepSeek at its peak rate.
Price
What the models cost, plus 5%.
The only fee. One balance for every model and every cloud.
See pricingDevelopers
Every task, on the best model for it.
Send Maestro the jobs from your own agent or app.
Apply for access