GPT-6 Astra is rolling out today to a limited set of organisations and will become available to all ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API and AWS, over the coming days.
It costs $10 per million input tokens and $50 per million output tokens, matching the price of Claude Fable 5 and 5.1. This is clearly OpenAI’s response to Fable. The model appears to score higher than Fable on most of OpenAI’s self-reported benchmarks.
Astra scores 99.9% on the ARC-AGI 3 benchmark, released in March. Fable 5 has not yet published a result. The ARC-AGI blog notes that the 99.9% score was achieved using OpenAI’s custom Provider Adapter harness for $19,000. The default ARC-AGI harness scored 62.7% for $26,000.
The Provider Adapter harness preserves opaque reasoning state between requests and uses compaction for longer conversations, allowing the model to reuse prior work.
Given the recent Hugging Face incident, Astra is a beast at security tasks. It scores 100% on ExploitBench, compared to GPT-5.6 Sol’s 78.5%. It scores 42.4% on ExploitGym, compared to Sol’s 30.3%. It achieves 99.2% within four attempts on SRE-Bench binary reverse engineering, compared to Sol’s 68.7%.
It is also better at long context. On OpenAI’s eight-needle benchmark it achieved 100% at 256,000 to 512,000 tokens and 96.3% at 512,000 to 1,000,000 tokens. OpenAI may have solved one of the ongoing challenges with long context processing.
It does not win at everything though. Artificial Analysis note that Astra is still beaten by Fable on their Intelligence Index.
Sits beside GPT-5.6 Sol in Intelligence: GPT-6 Astra scores equal to GPT-5.6 Sol in the Index at 61. This is five points lower than Claude Fable 5.1 (max with fallback). The model also trails Meta’s newly released Muse Spark 1.3 (max).
It performed better on their Coding Agent Index.
Leads Coding Agent Index cost efficiency frontier: At max effort, GPT-6 Astra costs about the same as GPT-5.6 Sol (max) while scoring two points higher on the Index. Per task, the model is less than half the cost of Claude Fable 5, for the same score.
I will write more about Astra once I get access to it. The API model label once it rolls out will be gpt-6-astra.
What it means
For developers and businesses, the main change is cost efficiency in coding tasks. Astra costs roughly the same as Sol at maximum effort but scores higher. It is less than half the cost of Fable 5 for the same score. Security teams face a new option for ExploitBench and SRE-Bench tasks, where Astra outperforms Sol significantly. Users with long documents or complex reasoning chains may find the 99.9% accuracy on ARC-AGI 3 useful, provided they use the custom harness that preserves reasoning state.
