GPT-6.2 Astra and Sol: Price, Context Window and What's New
GPT-6.2 Astra gets a 20% price cut to $8/$40 per million tokens, Sol holds steady at $2/$10, and both now ship with a 1.1M-token context window.
OpenAI released GPT-6.2 Astra and a new companion model, GPT-6.2 Sol, on November 17, 2026, cutting Astra's standard API price by 20 percent to $8 per million input tokens and $40 per million output tokens (down from $10/$50 for GPT-6 Astra), while Sol stays at $2 per million input tokens and $10 per million output tokens. Both models ship with a 1.1-million-token context window and roll out the same day across ChatGPT Work, Codex and the API.
The short version:
- GPT-6.2 Astra: $8 input / $40 output per million tokens, down 20% from GPT-6 Astra's $10/$50
- GPT-6.2 Sol: $2 input / $10 output per million tokens, unchanged from GPT-6.1 Sol
- Both models: roughly 1.1 million token context window, 128,000 token output limit
- Live now in ChatGPT Work, Codex and the API for Plus, Pro, Enterprise and Edu plans; not yet in standard ChatGPT chat
- API model IDs:
gpt-6.2-astraandgpt-6.2-sol
What OpenAI actually shipped on November 17
GPT-6.2 Astra and GPT-6.2 Sol are refreshed point releases that sit on top of OpenAI's existing GPT-6 line, rather than a brand-new generation. Astra remains the frontier model aimed at the hardest professional work — OpenAI's own framing points to documents, slides, specialized coding and research-grade tasks in fields like law, science and finance. Sol stays positioned as the everyday, high-throughput model meant to handle more routine and multistep agentic work for the same subscription price.
The headline change is price, not a new architecture announcement. Astra gets a straightforward discount across the board, while Sol's rates are left untouched from GPT-6.1 Sol, which launched in October. Coverage of the release, including a same-day write-up from AI news tracker TestingCatalog and a pricing breakdown from AI Weekly, frames the move as OpenAI making its top-tier model cheaper to run at scale rather than introducing new capabilities from scratch.
That framing matters for anyone budgeting API usage. A 20 percent cut on Astra's standard rate, plus a 60 percent cut on cached input, is the kind of change that shows up directly on an invoice for teams running high-volume coding agents or document-processing pipelines, even without any change in how the model behaves day to day. For consumer ChatGPT users, the more visible change is likely to be the context-window bump and whatever quality improvements OpenAI has quietly folded into the refreshed training run, since the subscription price itself hasn't moved.
GPT-6.2 Astra and Sol pricing: how much cheaper is it
Independent reporting on the API rate card agrees on the core numbers. GPT-6.2 Astra's standard input price drops from $10 to $8 per million tokens and its output price drops from $50 to $40 per million tokens — a 20 percent cut on both. Cached input pricing falls even further, from $1.00 to $0.40 per million tokens, a 60 percent reduction that matters most for applications that repeatedly reuse large prompts, such as coding agents working against the same codebase.
GPT-6.2 Sol is a straight carryover of GPT-6.1 Sol's pricing: $2 per million input tokens, $10 per million output tokens and $0.10 per million cached input tokens. OpenAI isn't discounting Sol further, because at $2/$10 it was already positioned as the budget option in the lineup; the news for Sol is the version bump and the shared context-window increase, not the price.
| Model | Input (per 1M tokens) | Cached input | Output (per 1M tokens) | Context window |
|---|---|---|---|---|
| GPT-6 Astra (Sept 2026) | $10.00 | $1.00 | $50.00 | ~1.05M tokens |
| GPT-6.2 Astra (Nov 2026) | $8.00 | $0.40 | $40.00 | ~1.1M tokens |
| GPT-6.1 Sol (Oct 2026) | $2.00 | $0.10 | $10.00 | ~1.05M tokens |
| GPT-6.2 Sol (Nov 2026) | $2.00 | $0.10 | $10.00 | ~1.1M tokens |
Two reports also detail pricing beyond the standard tier. Fast mode (previously called Priority processing) runs at double the standard rate on both models, and Ultrafast mode costs three times standard through February 2027, rising to six times standard from March 2027. For prompts longer than 272,000 input tokens, input and cached-input rates double and output is billed at 1.5x for the entire request — on Astra, that works out to $16 input, $0.80 cached and $60 output per million tokens once a prompt crosses that threshold.
Context window and technical specs
Both GPT-6.2 Astra and GPT-6.2 Sol carry a context window reported at roughly 1.1 million input tokens, with a 128,000-token output limit, matching the window OpenAI has used since GPT-6 Astra's original September launch. Both models accept text and image input and return text-only output; neither supports fine-tuning, and reasoning effort cannot be set to "none" on either model. Knowledge cutoffs are reported as May 2026 for Astra and June 2026 for Sol — several weeks later than their predecessors, consistent with a refreshed training run rather than a reused checkpoint.
On the API side, both models are available through Chat Completions, Responses, Realtime, Assistants and Batch endpoints, with streaming, function calling and structured outputs supported. The Responses API's tool set — web search, file search, code interpreter, hosted shell, computer use, MCP and tool search — carries over unchanged from the GPT-6 line.
Where you can use GPT-6.2 Astra and Sol today
Availability at launch is narrower than a full ChatGPT rollout. Both models are live day one in ChatGPT Work and in Codex, and through the API, for Plus, Pro, Enterprise and Edu users. According to reporting on the release, an OpenAI staff member confirmed the models have not yet been added to standard ChatGPT chat for all users — so if you don't see GPT-6.2 in your regular ChatGPT model picker yet, that tracks with what OpenAI has said so far. ChatGPT Business, along with the Free and Go tiers, wasn't mentioned in the announcement at all.
Ultrafast processing — OpenAI's fastest, most expensive serving tier — extends to Pro 500 subscribers for both models, while Pro 200 subscribers gain Ultrafast access for the first time, starting with Sol only. Coding agent platform Cognition added both models to Devin on day one, and the API model identifiers are gpt-6.2-astra and gpt-6.2-sol for anyone wiring them into existing integrations.
Benchmark and cost claims, as reported
OpenAI's own marketing claims that GPT-6.2 Astra costs less than a third per task compared with GPT-6 Astra on equivalent work, and third-party benchmark aggregator Artificial Analysis reportedly ranks Astra second on its Intelligence Index. On Cognition's FrontierCode 2.0 benchmark, Sol is said to finish within half a point of Astra while costing roughly a fifth as much — the kind of result that supports OpenAI's pitch for Sol as the practical default for everyday agentic work.
Other vendor and third-party figures circulating around the launch include a reported 100 percent score for Astra on ARC-AGI-3 (per ARC Prize) and a #7 ranking for Astra versus #9 for Sol on the independently run Vals Index, with Sol said to cost under a third as much per task. These are worth treating as vendor-reported or third-party-eval numbers rather than settled facts: outlets covering the release, including a report citing independent evaluations referenced by AI-focused outlet The Decoder, note that outside legal-task evals ranked Astra anywhere from 11th to 30th, and flagged hallucination rates in the 55-59 percent range on at least one test set for both new models. None of the cost-per-task methodology behind OpenAI's own claims has been published, so independent verification of the specific numbers is still pending.
Safety thresholds and new restrictions
Per OpenAI's Preparedness Framework, as reported in coverage of the system card accompanying this release, both GPT-6.2 Astra and GPT-6.2 Sol are rated at the Critical threshold for cybersecurity risk and High (just below Critical) for biological and chemical risk. Astra is also reported as the first model OpenAI has released externally that meets the High threshold for AI self-improvement; Sol does not meet that bar. That's a notable escalation from the GPT-6 Cyber security-focused model OpenAI shipped earlier in the GPT-6 cycle, and it comes with tighter cyber safeguards that reportedly also constrain some legitimate security research use cases. Reduced cyber refusals are limited to organizations OpenAI has approved for its "Daybreak Red" program, and sensitive biology-research access is now routed through a separate internal system OpenAI calls GPT-Rosalind, according to the reporting.
For a company that has spent much of 2026 fielding questions about how fast its models are improving relative to how well anyone — including OpenAI itself — can monitor what they're doing, clearing the AI self-improvement threshold for the first time with an externally released model is a milestone that's likely to draw as much scrutiny as the price cut does. Expect outside researchers to spend the weeks after launch probing exactly what that threshold means in practice, separate from OpenAI's own benchmark claims.
GPT-6.2 Astra vs Sol: which one should you actually use
The practical split hasn't changed much since OpenAI first drew this line with Astra and Sol back in October — see our earlier breakdown of how GPT-6.1 Sol and Astra differed — except the price gap between the two has narrowed slightly now that Astra is cheaper. At $8/$40 versus $2/$10 per million tokens, Astra still costs roughly four times as much on input and output, so it remains the model to reach for when a task genuinely needs the extra reasoning headroom: dense legal or scientific documents, demanding coding work, or anything where early testers cited in launch coverage — including Harvey, Legora and ClickUp — reported needing fewer tokens and fewer review rounds to get a usable result. One accounting-software partner, Basis, reportedly saw about 50 percent less time spent on long-horizon bookkeeping work using Astra.
Sol remains the better default for high-volume, everyday use: drafting, summarizing, routine coding, and multistep agent workflows where the cost difference compounds fast across thousands of calls. With FrontierCode 2.0 scores reportedly landing within half a point of Astra at a fraction of the cost, Sol looks like the model most teams should start with, upgrading to Astra only for the specific jobs that need it.
How GPT-6.2 fits into the rest of the GPT-6 family
This release doesn't replace the broader context OpenAI has built up around GPT-6 this year. Our original look at what GPT-6 Astra does and what it costs at its September launch is still useful background for understanding why Astra is priced the way it is, and how its capabilities differ from the smaller models in the lineup. For readers comparing OpenAI's models against the rest of the field, the cross-vendor rundown of how GPT-6 Astra stacks up against Gemini 4 Argon and Claude Opus 5.5 is a useful reference point, though it predates this pricing update and should be read with that in mind.
What hasn't been addressed yet, per the available reporting, is whether OpenAI plans to deprecate GPT-6 Astra and GPT-6.1 Sol now that their .2 successors exist, or on what timeline. OpenAI also hasn't said whether GPT-6.2 will reach standard ChatGPT chat, ChatGPT Business, or the Free and Go tiers, and it's unclear which regions get access to Ultrafast processing on the new models.
What to watch next
The things worth tracking from here: whether OpenAI widens GPT-6.2 Astra and Sol beyond ChatGPT Work and Codex into standard ChatGPT chat, whether the older GPT-6 Astra and GPT-6.1 Sol get formal deprecation dates, and whether independent benchmark groups publish eval results that confirm or complicate OpenAI's own cost-per-task claims. The gap between OpenAI's marketed benchmark wins and the more mixed results reported on specific legal and hallucination-focused evals is also worth watching — it's the kind of discrepancy that tends to get resolved, one way or the other, within the first few weeks after a model lands in more hands. As with past GPT-6 releases, expect OpenAI's own developer changelog and pricing page to be the definitive source once broader access rolls out; this article will be updated if the picture changes materially.
Frequently asked questions
What is GPT-6.2 Astra?
GPT-6.2 Astra is OpenAI's refreshed flagship model, released November 17, 2026, as an update to GPT-6 Astra. It targets the most demanding professional work (coding, documents, research) and costs $8 per million input tokens and $40 per million output tokens, a 20 percent cut from GPT-6 Astra's $10/$50 pricing.
What is GPT-6.2 Sol?
GPT-6.2 Sol is OpenAI's everyday, lower-cost model released alongside Astra on November 17, 2026. It's priced at $2 per million input tokens and $10 per million output tokens, unchanged from GPT-6.1 Sol, and is positioned for high-volume, routine and multistep agentic tasks.
How much do GPT-6.2 Astra and Sol cost?
GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens, with cached input at $0.40 per million tokens. GPT-6.2 Sol costs $2 per million input tokens, $10 per million output tokens, and $0.10 per million cached input tokens.
What is the context window for GPT-6.2 Astra and Sol?
Both models are reported to carry roughly a 1.1 million token context window with a 128,000 token output limit, in line with the window OpenAI has used across the GPT-6 family since GPT-6 Astra's original September 2026 launch.
Where can I access GPT-6.2 Astra and Sol?
Both models are available from launch in ChatGPT Work, Codex, and the API (as gpt-6.2-astra and gpt-6.2-sol) for Plus, Pro, Enterprise and Edu users. As of launch, OpenAI had not added them to standard ChatGPT chat, ChatGPT Business, or the Free and Go plans.
Is GPT-6.2 Astra cheaper than GPT-6 Astra?
Yes. GPT-6.2 Astra's standard input and output prices are both 20 percent lower than GPT-6 Astra's, dropping from $10/$50 to $8/$40 per million tokens, while cached input pricing fell 60 percent, from $1.00 to $0.40 per million tokens.
Sources
- TestingCatalog: OpenAI launches GPT-6.2 Astra and Sol on ChatGPT and APIstestingcatalog.com
- AI Weekly: OpenAI ships GPT-6.2 Astra at $8/$40 per million tokens, 20% below GPT-6 Astraaiweekly.co
- llm-stats.com: GPT-6.2 Astra model specsllm-stats.com
- datastudios.org: OpenAI GPT-6.2 Astra & Sol launch, API pricing, ChatGPT and Codexdatastudios.org
- The Decoder: OpenAI launches GPT-6.2 Astra and Sol with lower API pricesthe-decoder.com
Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

