Capability-based pricing. OSS rates by default.
Self-host the OSS engine for free — bring your own frontier keys or point your ChatGPT subscription at the gateway. Hosted is in invite-only beta; pricing opens with it.
Self-host ModelMeld. Bring your own frontier keys and your own GPU. Everything works on a snapshot of the routing data; the live curated feed is the upgrade.
- Self-host ModelMeld
- Bring your own frontier API keys
- Bring your own local GPU
- Bundled seed routing data (stale but functional)
- GitHub Issues + Discussions support
Pay-as-you-go across our routed model catalog. No DevOps, no GPU procurement, no key custody. Multi-model routing happens server-side; you point your dev tool at one base URL and we pick the right model per request. Currently in invite-only beta — the OSS engine is the path while we open it up.
- Multi-model routing across our hosted catalog
- BYOK passthrough — frontier keys transit per-request, never persisted
- Subscription passthrough — point your ChatGPT OAuth at us
- Pro routing feed bundled while you're an active customer
- Start in 5 minutes — no DevOps
Live curated routing feed for self-hosters who don't need our hosted compute. Routes against fresh benchmark data weekly. The OSS engine ships with a frozen snapshot today; Pro unlocks the auto-refreshed feed when it launches.
- Continuously curated routing feed
- Signed JSON delivery (Ed25519)
- Falls back to bundled snapshot on any failure
- Same OSS engine — self-host on your hardware
- Cancel anytime
Your hardware, your cloud, or hosted by us — pricing flexes to the deployment. SSO + RBAC, audit logs, custom SLAs, direct engineering support.
- Deploy on your hardware, your cloud, or ours
- SSO (OIDC) + RBAC
- SOC 2-grade immutable audit log
- Custom SLAs
- Direct engineering support
What hosted billing looks like when it opens.
Hosted will be pay-as-you-go via prepaid credits. Credits never expire. Optional auto top-up with a monthly cap — set it once and forget it. Top-up amounts and per-model rates land alongside your Hosted access invite.
What we route to.
The router picks per-request based on capability match. Frontier models bill against your own subscription or API key — no markup. Hosted-fleet models bill against your credit balance at the rate card.
| Model | Role | Billed against |
|---|---|---|
| claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5 | Frontier reasoning, coding, and tool-use — via your Anthropic API key | Your key |
| gpt-5, gpt-5-mini, gpt-5-4 | Frontier coding, simple-QA, agentic workflows — via ChatGPT subscription (Codex backend) or your OpenAI API key | Your key |
| gemini-3, gemini-2-5-pro | Long-context (2M tokens), reasoning — via your Google API key | Your key |
| qwen3-coder-next | OSS coding mid-tier (256k context) — tool-capable, weighted aggregator score on Aider Polyglot + SWE-Bench | Hosted credits |
| deepseek-v4-pro | OSS reasoning + coding — high tool-use score on internal evals | Hosted credits |
| kimi-k2.6 | OSS general-purpose reasoning at competitive cost; strong tool-use | Hosted credits |
| gpt-oss-120b | OSS mid-tier coding + reasoning — cheapest tool-capable option | Hosted credits |
| llama-3.3-70b-instruct | OSS general-purpose — reliable backstop when other OSS models are unhealthy | Hosted credits |
| llama-4-scout | OSS long-context (1M tokens) — for multi-file refactors and large-codebase requests | Hosted credits |
Catalog evolves as we add capacity and as models age out. The router prefers fresh, healthy capacity per request and falls through a capability-ranked chain if a model is unhealthy.
How the meter works
- →Hosted-routed requests: billed at the rate card above. Free if you self-host on your own hardware.
- →Frontier-routed requests:billed by the frontier provider directly, against the API key you provided. We never see the cost; we don't mark it up.
- →Routing decision itself: always free.
- →Auto top-up:set a monthly budget cap and we'll keep your balance funded inside it. Turn it off and you're back to manual top-ups.
- →No prompt-content storage: request metadata + SHA-256 prompt hash only.