Map Ox Alpha free-tier rate limits across providers #40

Open AtlantisPleb opened this 5h ago 1 comment

Ox Alpha is free for roughly four days through OpenRouter, OpenCode (100 trillion tokens per day of advertised capacity), and the Nous Research Portal (advertised 1 quadrillion tokens/day). Probe each provider to find real rate limits: concurrent request ceilings, TPM/RPM caps, context limits, and whether limits are per key or per organization. Output a table of measured limits so we can split workloads across all providers instead of guessing.

AtlantisPleb opened this issue 5h ago
A AtlantisPleb Author 4h ago
Measured so far:
- Nous Portal model API: verified. GET https://inference-api.nousresearch.com/v1/models (public, no auth) lists stealth/ox-alpha at $0/$0, context 1,048,576, max output 131,072, reasoning mandatory with default effort max. OpenAI-compatible chat completions with tools and response_format. Tier limits from the Portal docs: Free 50 RPM / 500K TPM; Plus 400 RPM / 4M TPM; Super 800 RPM / 8M TPM; Ultra 1600 RPM / 16M TPM. Even Ultra is ~23B tokens/day, three orders of magnitude under opencode's claim.
- Hermes Agent carries Ox Alpha but routes through opencode and OpenRouter per the announcement, so it is a wrapper, not an extra lane.
- OpenRouter: free via stealth/ox-alpha, one stealth provider, community-observed ~18-26 TPS.
- opencode: claimed 100T tokens/day capacity (~1.16G TPM equivalent). Unverified empirically.
- Venice: still unverified for ox-alpha.
Remaining work: probe each lane's real concurrent-request ceiling under load, then record per-lane budgets in the overflow router issue #56.