OpenAI's GPT-6 Stack: Astra Flagship, Plus Sol and Luna

OpenAI introduced GPT-6 Astra as its flagship model and, in a Sep 22, 2026 update on the same page, said it is expanding the GPT-6 family with GPT-6 Sol and GPT-6 Luna. Primary source: OpenAI — GPT-6 Astra.

What shipped (claimed)

• GPT-6 Astra: flagship; vendor claims SOTA on computer use, browsing, software engineering, cybersecurity, science, and professional work
• Sep 22 update: GPT-6 family expands beyond Astra with Sol and Luna
• Rolling out to ChatGPT Plus / Pro / Business / Enterprise; API, Azure, and Bedrock; Enterprise Astra off by default

Caveat: Benchmarks and capability claims below are vendor / eval-environment claims from OpenAI's page — not independent bake-offs.

Benchmark snapshots vs GPT-5.6 Sol (vendor)

• OSWorld 2.0: 72.6% vs 65.7%, at about 47% less time per task
• Terminal-Bench 4.0: 57.9% vs 37.3%
• ExploitBench: 100% vs 78.5%
• ARC-AGI-3: 99.9% vs 7.8%

Price and availability

• API Standard: $10/M input, $50/M output
• Fast mode: up to 2x speed at 2x Standard price
• Also on Azure and AWS Bedrock; ChatGPT Plus/Pro/Business/Enterprise rollout

Cyber posture (claimed)

• Meets Critical threshold under OpenAI's Preparedness Framework for cybersecurity
• Stronger safeguards at launch; more advanced exploit-style tasks refused in the shipping version
• Daybreak planned later for less restrictive defender access

Operator takeaway

Treat Astra / Sol / Luna as a tiered stack, not a single seat upgrade. Route by finished-job cost and task difficulty — do not send every seat to flagship. Map work into leave / augment / automate bundles before you expand quota. Soft next step if useful: an Ada conversation to rank which task bundles belong on Astra vs Sol/Luna once pricing and latency land in your account.

Source: OpenAI — GPT-6 Astra: A new generation of intelligence (Sep 22, 2026 update).