NEWS →  ·  Models · OpenAI

OpenAI launches GPT-6.2 Astra and Sol: frontier API price cut 20%, live in Work, Codex and the API today

Nov 17, 2026

OpenAI released two models today, Nov 17, 2026: GPT-6.2 Astra, its new frontier model for the hardest projects, and GPT-6.2 Sol, an efficient everyday model that nearly matches Astra on real-world coding at about a fifth of the cost. Both are live in ChatGPT Work and Codex and in the API. Astra's API price falls 20% on its predecessor, to $8 per million input tokens and $40 per million output; Sol holds its predecessor's $2 and $10. Both models are closed — there are no weights to rent a GPU for — so this launch matters as the price bar every self-hosting decision is measured against.

OpenAI launches GPT-6.2 Astra and Sol: frontier API price cut 20%, live in Work, Codex and the API today

Image: OpenAI via community.openai.com

What OpenAI announced, from its own announcement post published Nov 17, 2026 and read first-hand (with the launch thread on X at 17:54 UTC): GPT-6.2 Astra and Sol are available starting today in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and to developers in the API as gpt-6.2-astra and gpt-6.2-sol, with the high-speed Ultrafast mode available — Pro 500 subscribers get Ultrafast with either model, and Pro 200 subscribers get it for the first time, starting with Sol. The rollout does not add either model to regular Chat. Astra, OpenAI says, sets new highs on evaluations across professional work and scientific research at less than a third of the cost per task of GPT-6 Astra; on Agents' Last Exam, GPT-6.2 Astra at extra-high effort scores 5.2 points above GPT-6 Astra at maximum effort at less than half the cost per task. Astra is also OpenAI's most capable model yet for cybersecurity, jumping 22 points over GPT-6 Astra on its ExploitBench Internal Port evaluation — and because of that capability, both models launch with tighter safeguards for higher-risk cybersecurity assistance, with work under way to give approved defenders reduced-refusal access through Daybreak. All benchmark figures are OpenAI's own.

The pricing is the part builders will price against. Astra costs $8.00 per million input tokens and $40.00 per million output tokens in the API — a 20% cut from GPT-6 Astra's $10 and $50, with cached input down 60% to $0.40 per million on OpenAI's API documentation. Sol keeps GPT-6.1 Sol's rates exactly: $2.00 in, $0.10 cached in, $10.00 out per million. OpenAI's claim for Sol is that it nearly matches Astra where it counts for agent work: on FrontierCode 2.0, which tests merge-ready changes to real codebases, Sol finishes within about half a point of Astra at roughly one-fifth of the cost. Independent scoring is already landing: Artificial Analysis puts Astra at 62.1 and Sol at 57.3 on its Intelligence Index. Neither model is open-weight — no weights, no licence, no self-hosting route was announced.

What it means if you rent GPUs

For GPU renters and model builders, a closed launch like this changes no board price and adds nothing to the catalogue — but it moves the number every deployment plan is compared against. The frontier API just got 20% cheaper per token, and on OpenAI's own figures the cost per finished task fell by half or more, which pushes the break-even usage level for renting GPUs and self-hosting an open-weight model higher: you now need more volume, or a stronger privacy and control case, before the rented route wins on cost alone. Sol holding $2 and $10 while closing most of the gap to the frontier model is the volume story — that is the rate high-throughput agent workloads will be benchmarked against. We will keep tracking the open-weight answers to this pair in the catalogue, priced from the board we check every day.

Go deeper

← All news