gpt-6-astra is live in both the default and svip groups
OpenAI’s next-generation flagship, released on 3 September and positioned for computer use, software engineering, and long-horizon agent work. Specs: 1,050,000-token context, 128,000-token max output, knowledge cutoff 30 April 2026; reasoning_effort gains xhigh / max for five levels in total.
Pricing matches OpenAI’s standard tier item for item: $10 in / $50 out per 1M tokens, $1 cached reads, $12.50 cache writes. That is 2.5x the current promotional price of gpt-5.6-sol and identical to Claude Fable 5.1 on input and output. The /v1/chat/completions endpoint (OpenAI-compatible format) is open now; migrating from 5.6 Sol is a model-field change.
Two notes: OpenAI rates Astra’s cyber capability at the Critical level, so the standard public release refuses vulnerability-discovery and exploit-writing tasks, while ordinary development is unaffected; billing is tiered by input tokens exactly as on OpenAI, and once a single request’s input passes 272K the whole request bills at the second tier ($20 input, $2 cached read, $25 cache write); treat the live model pricing page as authoritative.
📖 Benchmarks, model selection, and migration notes: GPT-6 Astra Is Live: OpenAI’s New Flagship Open for Calls
← Back to Live Updates · 📚 Monthly archive