GPT‑6 Astra

OpenAI is rolling out GPT-6 Astra to a limited set of organizations, with broader availability coming for ChatGPT paid tier users, enterprise customers, the OpenAI API, and AWS. The model is priced through the API at $10 per million input tokens and $50 per million output tokens, positioning it as OpenAI's competitor to Claude Fable.

On self-reported benchmarks, Astra outperforms Fable and achieves a 99.9 percent score on the ARC-AGI 3 benchmark using a custom Provider Adapter harness that preserves reasoning state and uses conversation compaction. However, the default harness for the same benchmark scored 62.7 percent. Astra also excels in security tasks, scoring 100 percent on ExploitBench, 42.4 percent on ExploitGym, and 99.2 percent within four attempts on SRE-Bench binary reverse engineering. In long context testing, it achieved 100 percent on the eight-needle benchmark at 256K to 512K tokens and 96.3 percent at 512K to 1M tokens.

Despite these high scores, Astra does not win in every category. Artificial Analysis places Astra equal to GPT-5.6 Sol on its Intelligence Index at a score of 61, which is lower than Claude Fable 5.1 and Meta's Muse Spark 1.3. However, Astra leads the Coding Agent Index cost efficiency frontier, scoring slightly higher than GPT-5.6 Sol at a similar max-effort cost, and achieving the same score as Claude Fable 5 at less than half the per-task cost. The API model label will be gpt-6-astra.