Sign in with GitHub lands alongside EvalLab — a tool for running model comparisons — plus a full speech-to-text interface, low balance email alerts, and new providers.
GitHub sign-in
Sign in and link your account with GitHub OAuth, alongside the existing Google and Apple options.
EvalLab
A new EvalLab section in the dashboard lets you create evaluation suites, define test cases with multiple scoring methods, and run them across any set of models simultaneously. Results show per-model scores, test-case breakdowns, and cost estimates — so you can make data-driven model choices instead of guessing.
Speech-to-text
A dedicated transcription page is now available in the dashboard — upload an audio file or record directly in the browser and get a transcript back. The audio endpoint now also exposes Cartesia and AssemblyAI as STT providers alongside the existing options, covering a wide range of audio/video formats.
Low balance alerts
Set a credit threshold and add up to 10 email recipients — when your balance drops below the threshold, everyone on the list gets notified. No more unexpected outages from a drained account.
New models via Cerebras
– Llama 3.1 8B — 32K context, streaming, tools, structured output.
– Qwen 3 235B Instruct — 131K context, fully featured.
– Zai GLM 4.7 — 131K context.
– GPT-OSS 120B on Cerebras — High-throughput variant via the Cerebras provider.