Blog
Benchmarks, migration notes and how the service is run. RSS
We captured every request Codex CLI 0.154.0 makes when pointed at a custom provider: a 17,000-character system prompt on every turn, 17,500 characters of tool definitions, a stable installation id, and two request shapes that only OpenAI understands. Here is what is in them, why MCP tools and sub-agents break on open models, and what a provider can do about it.
Whisper lists 100 languages. We sent 30 public test clips per language through our API and scored every transcript. Around a third of languages are good or excellent, a third are not usable, a few come back empty, and auto-detection confuses more languages than you would expect.
We timed every chat model on the public endpoint with a fixed prompt: time to the first token, how long each model thinks before the visible answer starts, and output tokens per second. Includes a correction to the figures we first published.
On 16 August 2026 Groq shut down llama-3.3-70b-versatile and llama-3.1-8b-instant for free and developer tiers. A month later, hundreds of thousands of code files still pin the id. What the error looks like, who it hit, and the fix ladder.
Every model is priced at about 30% over what it costs us to serve, and the rate is on the model page. What that means in practice, and why we do not have tiers.