Update: GPT-6 Astra Pro adds a pro-reasoning dial on OpenRouter
Same weights as Astra, one flag flipped: reasoning.mode=pro. Price holds at $10/$50 per M — the dial just burns more thinking tokens on your hardest jobs.
Copy markdown
Same model, a reasoning dial
GPT-6 Astra Pro isn't new weights — it's GPT-6 Astra served with reasoning.mode set to pro. Flip it on for the problems base Astra fumbles, leave it off for everything else, and you're running the identical model underneath.
Same price, higher task cost
Input is $10/M and output $50/M — identical to base Astra. Pro doesn't raise the sticker; it spends more reasoning tokens per answer, so budget by the task, not the token, and expect the hard problems to cost more to finish.
1M context, 128K output intact
Pro inherits Astra's 1.05M-token context window and 128K max output. Whole-repo reviews and giant transcripts work exactly as before — the dial changes how hard the model thinks, not how much it can read or write.
Pick your provider carefully
OpenRouter fans Astra Pro across multiple hosts, with latency from 2 to 29 seconds and throughput of 19 to 97 tokens/sec. Pro reasoning is slower by design, so route latency-sensitive calls to base Astra or a fast host and save Pro for jobs that can wait.