OpenAI announced GPT‑6 Astra, a new language model rolling out today to a limited set of organizations and, over the coming days, becoming available to all ChatGPT Plus, Pro, Business, and Enterprise users. The model will also be accessible through the OpenAI API and on Amazon Web Services (AWS). The author of the report notes they have not yet tried the model personally.
Availability and pricing
The API pricing for GPT‑6 Astra matches Claude Fable 5 and 5.1: $10 per million input tokens and $50 per million output tokens. When it becomes generally available via API, its model label will be gpt-6-astra.
Performance and benchmarks
OpenAI and third‑party benchmarks published so far show strong performance for Astra on multiple fronts.
-
ARC‑AGI 3: Astra scored 99.9% on the ARC‑AGI 3 benchmark (released in March). OpenAI’s blog notes that the 99.9% result used a custom "Provider Adapter harness" costing $19,000; by contrast the default ARC‑AGI harness scored 62.7% at a reported cost of $26,000. The Provider Adapter harness preserves an opaque reasoning state between requests and uses compaction for longer conversations, allowing the model to reuse prior work.
-
Security tasks: Astra performs exceptionally on security assessments — 100% on ExploitBench (compared with GPT‑5.6 Sol’s 78.5%), 42.4% on ExploitGym (Sol 30.3%), and 99.2% within four attempts on SRE‑Bench binary reverse engineering (Sol 68.7%). The article highlights these results as particularly relevant following the recent Hugging Face incident.
-
Long‑context handling: On OpenAI’s eight‑needle benchmark, Astra achieved 100% at 256K–512K tokens and 96.3% at 512K–1M tokens, indicating substantial improvements in processing very long contexts.
Areas where Astra trails competitors
Astra is not the leader on every independent metric. Artificial Analysis reports GPT‑6 Astra at 61 on its Intelligence Index, equal to GPT‑5.6 Sol but five points below Claude Fable 5.1 (max). Astra also trails Meta’s newly released Muse Spark 1.3 (max) on that index.
On the Coding Agent Index, Astra performed well: at maximum effort its cost is roughly the same as GPT‑5.6 Sol (max) while scoring two points higher on the Index. Per task, Astra’s cost is less than half that of Claude Fable 5 for a comparable score.
Summary
GPT‑6 Astra positions OpenAI as a direct competitor to Claude Fable, matching Fable’s API pricing while delivering strong results on several benchmarks — notably security tests and long‑context processing. However, on some independent intelligence measures it does not lead the field. The author plans to publish further analysis after gaining hands‑on access.
Practical note
As the rollout completes, developers and enterprises will be able to compare gpt-6-astra directly against other available models based on the published benchmarks and API pricing.
