Project Ares API Platform — every frontier model, one key

1

endpoint — POST /v1/messages, streaming passed through byte-for-byte

8+

providers fronted — Claude, GPT, and the best open models behind one key

30+

models in the living registry, with lifecycle metadata and retirement dates

90%

of input cost saved on cached prompt prefixes — reads bill at 10% of input

01Why build on Project Ares

The parts of shipping with LLMs nobody wants to own

Provider accounts, model shutdowns, invoices from four vendors, spend nobody can explain. We run that layer so your team ships product.

⬡

One endpoint, every model

POST /v1/messages with streaming passed through byte-for-byte. Keep the SDK you already use — just change the base URL — and reach Claude, GPT, and the best open models through the same request shape.

◈

A living model registry

GET /v1/models lists every model across every provider with lifecycle metadata. New models appear automatically; deprecated ones carry retirement dates and a replacement; retired ones fail loudly with the fix in the error.

⚠

Deprecation warnings built in

Calling a deprecated model returns an x-ares-deprecation header with the sunset date and suggested replacement — your logs tell you about migrations before the provider breaks you.

▤

One bill, real controls

Prepaid credits with auto-reload, monthly spend limits per account, and a ledger you can audit. Every request is metered per token and priced transparently.

◫

Organizations & roles

Owners, admins, members. Per-member usage insight, single-use email invites, keys that never bleed across accounts.

⛨

Security by default

Keys hashed and shown once, CSRF-guarded dashboard, login lockouts, audit trail, anomaly flags. Provider credentials never leave our gateway.

02The living registry

Models retire. Your app shouldn't notice.

Every model in the registry carries lifecycle metadata, and the platform walks you through the whole arc — keep scrolling to watch one model age gracefully instead of breaking you at 2 a.m.

Today · Active

Ship on the best model

GET /v1/models lists everything across every provider. New models appear automatically — no SDK update, no announcement thread to catch.

"id": "claude-sonnet-4-6" · "status": "active"

Later · Deprecated

The warning comes to you

Keep calling it — every response now carries a header with the sunset date and the suggested replacement. Your own logs announce the migration.

x-ares-deprecation: sunset=2027-03-01; replacement=claude-sonnet-5

Eventually · Retired

Failure with the fix inside

Retired models fail loudly — and the error tells you exactly what to change. One string. That's the whole migration.

410 model_retired · "use claude-sonnet-5 — same request shape"

03Getting started

First request in under two minutes

No SDK changes if you already use Anthropic's — one base URL swap and you have every model.

Create your account

Sign up and open the API Platform Console — free credits are included, no card needed.

api.projectares.ai/dev

Create an API key

Console → API keys → Create. It's shown once and stored hashed.

sk-ares-••••••••••••

Call any model

curl, or your existing SDK with baseURL pointed at us. Pick a model — the request doesn't change:

curl https://api.projectares.ai/v1/messages \
  -H "x-api-key: $ARES_KEY" \
  -d '{"model":"claude-sonnet-5",
       "max_tokens":300,
       "messages":[{"role":"user",
         "content":"hello"}]}'

Get your API key Full API reference Try the playground

04Model pricing

Every model, priced in the open

Simple per-token pricing in dollars, metered exactly. Pick a brand to see its models — the full catalog (and every dated variant) is on GET /pricing.

Model	Input / MTok	Output / MTok	Cache read	Cache write	Priority (opt-in)
Loading live pricing…

All prices are USD per million tokens (MTok). Live from GET /pricing — the same table the meter bills from, so this page can never disagree with your ledger.

Prompt caching

Reuse your prompt, keep 90% of the cost

When your requests share a long prefix — a system prompt, a document, tool definitions — the model can cache it. Cached tokens bill at just 10% of the input price on every subsequent request; writing the cache costs 1.25× input (2× for 1-hour caches). On models marked automatic, it just happens — no code changes. For agents and RAG apps this routinely cuts input spend by 80–90%.

Priority service

A fast lane when latency matters

Send speed: "fast" and supported models are served from priority capacity — faster first token and steadier throughput at peak. The trade is price, not quality: fast requests bill at 2× the token prices shown above (the "2× price" in the table). Only requests actually served fast are billed fast — if a model doesn't support it, the request simply runs and bills at standard. No config, no commitments — flip it per request.

05One sign-in, two products

Same account. Separate bills.

Your Project Ares sign-in opens both the Ares agent and the API Platform — but they are separate products with separate billing: each keeps its own balance, its own plan, and its own invoices. API spend never mixes with your agent usage.

Meet Ares — the agent →

Every frontier model.
One key. One bill.