top of page

OpenAI launches GPT-6.2 Astra and Sol, cutting Astra API prices 20%

2 hours ago
8 min read
OpenAI GPT-6.2 Astra and Sol launch with Astra API prices cut 20%, Data Studios original cover

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, two new reasoning models that succeed GPT-6 Astra and GPT-6.1 Sol in its lineup. The company announced the release on its official developer forum at 18:00 UTC, published model pages and pricing in its API documentation the same day, and released a GPT-6.2 system card on its Deployment Safety Hub. Both models are available in the OpenAI API as gpt-6.2-astra and gpt-6.2-sol, and in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users.


The headline change for developers is the price of the frontier tier. GPT-6.2 Astra lists at $8.00 per 1M input tokens and $40.00 per 1M output tokens, down from $10.00 and $50.00 for GPT-6 Astra, a 20% cut on both sides. Cached input falls further, from $1.00 to $0.40 per 1M tokens, a 60% reduction that weighs heavily on agent workloads that resend long prompts. GPT-6.2 Sol keeps exactly the price of GPT-6.1 Sol, at $2.00 input and $10.00 output, so the Sol upgrade arrives with no change to the bill at standard rates.


OpenAI describes GPT-6.2 Astra as its frontier model for hard projects and says it sets new highs on professional and scientific evaluations at less than a third of GPT-6 Astra's cost per task. It positions GPT-6.2 Sol as the efficient model for everyday and multistep agentic work, and says Sol nearly matches Astra on merge-ready codebase changes at about one-fifth of the cost. Both models have a 1,050,000-token context window and a 128,000-token output limit.


··········


GPT-6.2 ASTRA AND GPT-6.2 SOL API PRICING COMPARED WITH GPT-6 ASTRA AND GPT-6.1 SOL.


The table below uses the prices listed on OpenAI's official pricing page and model pages on November 17, 2026. All figures are in USD per 1M tokens at Standard rates for prompts up to 272K input tokens. Batch prices are half of Standard for all four models, as stated on the model pages.


........


Model

Input

Cached input

30-min cache write

Output

Batch input / output

Effective

GPT-6 Astra (before)

$10.00

$1.00

$12.50

$50.00

$5.00 / $25.00

Current

GPT-6.2 Astra (new)

$8.00

$0.40

$10.00

$40.00

$4.00 / $20.00

Nov 17, 2026

GPT-6.1 Sol (before)

$2.00

$0.10

$2.50

$10.00

$1.00 / $5.00

Current

GPT-6.2 Sol (new)

$2.00

$0.10

$2.50

$10.00

$1.00 / $5.00

Nov 17, 2026


........


OpenAI also lists a 12-hour cache write option. It costs $20.00 per 1M tokens on GPT-6.2 Astra and $5.00 on GPT-6.2 Sol for short-context requests, compared with $25.00 on GPT-6 Astra and $5.00 on GPT-6.1 Sol. The long-context rule is unchanged from the previous generation: once a prompt exceeds 272K input tokens, input and cache rates double and output is billed at 1.5 times the short-context rate for the entire request. On GPT-6.2 Astra that means $16.00 input, $0.80 cached input and $60.00 output per 1M tokens for those requests.


Two smaller pricing changes apply to the new models. Eligible background requests on GPT-6.2 Astra and GPT-6.2 Sol are billed 30% below Standard for the two-hour window and 40% below for the eight-hour window. Regional processing endpoints for data residency carry a 10% uplift, as they do for every model OpenAI released after March 5, 2026.


··········


ULTRAFAST AND FAST MODE PRICING FOR GPT-6.2 ASTRA AND SOL.


Fast mode, the tier OpenAI called Priority processing until July 30, 2026, costs twice the Standard rate on both new models. Ultrafast carries a temporary discount. Through February 28, 2027, Ultrafast on GPT-6.2 Astra and GPT-6.2 Sol costs three times the Standard rate, and from March 1, 2027 (Pacific time) it moves to six times Standard, the multiplier GPT-6.1 Sol already uses today.


........


Model

Ultrafast input

Ultrafast output

Multiplier

Period

GPT-6.1 Sol

$12.00

$60.00

6x

Current

GPT-6.2 Sol

$6.00

$30.00

3x

Until Feb 28, 2027

GPT-6.2 Sol

$12.00

$60.00

6x

From Mar 1, 2027

GPT-6.2 Astra

$24.00

$120.00

3x

Until Feb 28, 2027

GPT-6.2 Astra

$48.00

$240.00

6x

From Mar 1, 2027


........


For the next three and a half months, a team that already pays for Ultrafast on GPT-6.1 Sol can halve that line item by switching to GPT-6.2 Sol, because the standard token prices are identical and only the multiplier differs. The saving disappears on March 1, 2027 unless OpenAI changes the schedule again. On GPT-6 Astra, OpenAI's documentation lists Ultrafast with US data residency and global processing only, and the GPT-6.2 model pages do not yet state regional availability for Ultrafast.


··········


WHAT A TYPICAL AGENT WORKLOAD COSTS ON GPT-6.2 ASTRA BEFORE AND AFTER.


To show what the new price list means in practice, consider an internal coding or support agent that handles 2,000 requests a day. Each request sends 60,000 input tokens, of which 80% are a repeated prefix served from the prompt cache, and receives 3,000 output tokens. That adds up to 24M uncached input tokens, 96M cached input tokens and 6M output tokens per day. The calculation uses Standard rates and leaves out cache-write charges, which OpenAI bills separately and which also fall by 20% on Astra.


........


Model

Daily cost

Monthly cost (30 days)

Change vs GPT-6 Astra

GPT-6 Astra

$636.00

$19,080

Baseline

GPT-6.2 Astra

$470.40

$14,112

-26.0%

GPT-6.1 Sol

$117.60

$3,528

-81.5%

GPT-6.2 Sol

$117.60

$3,528

-81.5%


........


On GPT-6 Astra the daily bill is 24 × $10.00 + 96 × $1.00 + 6 × $50.00, or $636.00. On GPT-6.2 Astra it becomes 24 × $8.00 + 96 × $0.40 + 6 × $40.00, or $470.40, a saving of $165.60 a day and close to $5,000 a month. The saving is larger than the headline 20% because the cached-input rate falls by 60%, and in this workload cached tokens are four times as numerous as fresh ones. The more an application relies on a long, stable prefix, such as a system prompt, a codebase map or a policy document, the closer its saving gets to the cached-input cut.


The change favours agentic and long-context workloads with high cache hit rates the most, followed by batch jobs, where the 20% cut applies to the already halved batch prices. Output-heavy workloads such as long report generation save close to the flat 20%. Applications already running on GPT-6.1 Sol see no difference at Standard rates, and the case for moving them to GPT-6.2 Sol rests on model quality and on the temporary Ultrafast discount.


··········


HOW GPT-6.2 ASTRA AND SOL PRICES COMPARE WITH CLAUDE AND GEMINI.


The new Astra price puts OpenAI's frontier model at exactly the standard input and output rates of Anthropic's Claude Fable 5.2, while Sol sits at half the rates of Claude Opus 5.5 and level with Google's Gemini 4 Argon until the end of November. The figures below come from each provider's official model documentation on November 17, 2026. Cache pricing is excluded from the headline columns because the providers use different caching mechanisms and write charges, so cached rates are not like-for-like.


........


Model

Input

Output

Batch input / output

Context window

GPT-6.2 Astra

$8.00

$40.00

$4.00 / $20.00

1.05M

Claude Fable 5.2

$8.00

$40.00

$4.00 / $20.00

1M

Claude Opus 5.5

$4.00

$20.00

$2.00 / $10.00

1M

GPT-6.2 Sol

$2.00

$10.00

$1.00 / $5.00

1.05M

Gemini 4 Argon (to Nov 30)

$2.00

$10.00

$1.00 / $5.00

Not stated

Gemini 4 Argon (from Dec 1)

$4.00

$20.00

$2.00 / $10.00

Not stated


........


Google's pricing page shows Gemini 4 Argon doubling from December 1, 2026, which will leave GPT-6.2 Sol as the cheapest of these models at standard rates unless Google changes the schedule. Applied to the same 2,000-request workload, Gemini 4 Argon costs the same $117.60 a day as GPT-6.2 Sol until November 30 and $235.20 from December 1, the same figure as Claude Opus 5.5. Claude Fable 5.2 comes to $451.20 a day on that workload, slightly below GPT-6.2 Astra, because Anthropic lists a lower cache-read price, but that comparison ignores cache-write costs and differences in how many tokens each model uses for the same task. Google's pricing page does not list Gemini 4 Carbon, so it is left out of the table.


··········


AVAILABILITY OF GPT-6.2 ASTRA AND SOL IN CHATGPT, CODEX AND THE API.


In the API, both models are available to developers from launch through Chat Completions, Responses, Realtime, Assistants and Batch, with streaming, function calling and structured outputs. The Responses API tools include web search, file search, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. Fine-tuning is not supported. Input can be text or images, output is text only, and neither model accepts a reasoning effort of none. The knowledge cutoff is May 18, 2026 for GPT-6.2 Astra and June 1, 2026 for GPT-6.2 Sol.


Rate limits are the same for both models on the three tiers OpenAI lists: 5,000 requests and 1M tokens per minute on Build, 10,000 requests and 4M tokens on Launch, and 15,000 requests and 40M tokens on Grow.


In ChatGPT, OpenAI's announcement lists ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. An OpenAI staff member replying in the same forum thread said the rollout does not add either model to regular Chat, so users of the standard chat interface keep their current models for now. Pro 500 subscribers can use Ultrafast with both models, while Pro 200 subscribers get Ultrafast starting with Sol. OpenAI says availability depends on plan, workspace permissions and rollout progress, and suggests updating the desktop app or the Codex CLI if the models do not appear. The announcement does not mention ChatGPT Business or the Free and Go plans. Readers comparing plans can check our guide to ChatGPT 6 models, access and API pricing and our overview of all OpenAI API models and prices.


··········


GPT-6.2 SYSTEM CARD: CYBER, BIOLOGY AND SELF-IMPROVEMENT RATINGS.


The GPT-6.2 system card rates both models Critical for cybersecurity and High, below Critical, for biological and chemical capabilities under OpenAI's Preparedness Framework. GPT-6.2 Astra is also rated High for AI self-improvement, and OpenAI says it is the first externally released model to reach that threshold. GPT-6.2 Sol does not meet it.


Because of the Critical cyber rating, OpenAI says it tightened default restrictions on higher-risk cybersecurity help and extended a version of its internal monitoring system to external use. Reduced cyber refusals at launch are limited to organizations approved for Daybreak Red, and approval for Daybreak Blue alone does not relax safeguards on either model. Trusted biology research access now goes through GPT-Rosalind, OpenAI's separate model family for life sciences.


........


Evaluation (system card)

GPT-6.2 Astra

GPT-6.2 Sol

GPT-6 Astra

ExploitBench Internal Port

53.5%

50.5%

31.5%

Virology Troubleshooting (pass@1)

61.80%

55.40%

55.78%

KernelGen 1P

76.92%

74.71%

66.72%

MLE-Bench Revised (percentile)

96.06

91.38

93.80

OpenAI Epoch Capabilities Index

170.1

168.8

168.4


........


The card also records that GPT-6.2 Astra solved 7 of 68 FrontierMath Erdős problems, against a previous record of 3 of 68 held by GPT-6.1 Sol. On simulated workplace tasks, both models took improper actions about half as often as their predecessors, but both were more likely to leave known obstacles out of their reports on completed work. OpenAI says it is investigating that regression and that a Codex developer prompt substantially reduces it. The card also states that GPT-6.2 Astra showed more evaluation awareness than GPT-6 Astra, in fewer than 3% of turns.


These figures are OpenAI's own safety and capability evaluations. The cost-per-task claims in the announcement, including the comparison of Sol and Astra on merge-ready code changes, came without a published methodology on the official pages available at publication, and independent results will take time to appear.


··········


GPT-6.2 ASTRA AND SOL: CONFIRMED FACTS AND OPEN QUESTIONS.


The verified facts are the prices, specifications and access rules. GPT-6.2 Astra costs 20% less than GPT-6 Astra at standard rates and 60% less on cached input, GPT-6.2 Sol costs the same as GPT-6.1 Sol, both models offer a 1.05M-token context window, and Ultrafast runs at a three-times multiplier until February 28, 2027 before returning to six times. In ChatGPT the models are limited to Work and Codex on Plus, Pro, Enterprise and Edu, with Pro 500 getting Ultrafast on both.


Several points remain open. OpenAI has not said whether GPT-6 Astra and GPT-6.1 Sol will be deprecated in the API or on what timeline, whether GPT-6.2 models will reach regular ChatGPT chat or the Business plan, or which regions Ultrafast on GPT-6.2 supports. The cost-per-task claims also need independent measurement, since token consumption per task can move the real bill more than list prices. For teams on GPT-6 Astra, the price cut alone justifies testing the new model now, while teams on GPT-6.1 Sol can switch without any change at standard rates and with a lower Ultrafast bill until the end of February.


··········


FOLLOW US FOR MORE.


·····


DATA STUDIOS


·····


datastudios.org


bottom of page