OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, splitting its latest model release between difficult professional projects and efficient everyday work. Both models are available in ChatGPT Work, Codex and the API.
Astra carries the headline price change: input and output tokens cost 20% less than GPT-6 Astra, while cached-input tokens cost 60% less. Sol retains GPT-6.1 Sol’s rates, putting it at one-quarter of Astra’s Standard token prices.
The practical distinction is between token pricing and the cost of completing useful work. OpenAI reports improvements in both, but only the published token-rate reductions are straightforward to calculate. Per-task savings depend on how much reasoning, output and tool use a particular job requires.
Astra and Sol Target Different Workloads
OpenAI positions GPT-6.2 Astra as its frontier model for complex projects, including software development, scientific work, slides, spreadsheets and documents. The company says Astra follows professional conventions and supplied templates more closely, produces more direct writing, and improves coding and frontend development.
GPT-6.2 Sol targets general-purpose work, including writing, technical assistance and multistep agent workflows. OpenAI presents it as the efficient default, with Astra available for more demanding assignments. Those quality descriptions are vendor claims, not findings from hands-on testing for this article.
The launch provides these access options:
- ChatGPT Work and Codex: Both models are available to Plus, Pro, Enterprise and Edu users. Eligible enterprise and education access remains subject to workspace controls.
- OpenAI API: Developers select
gpt-6.2-astraorgpt-6.2-sol. - Ultrafast: Available for both models at launch, with additional pricing considerations.
I’d treat Sol as the starting candidate for routine automation because its published token rates are substantially lower. Astra earns its premium when better results, fewer retries or less downstream editing offset that difference. The announcement’s benchmarks suggest where to investigate that tradeoff; they cannot settle it for every workload.
API Pricing: What the Discounts Actually Save
OpenAI’s API pricing documentation lists the following Standard, short-context rates, in US dollars per million tokens:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $0.40 | $40.00 |
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.2 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
Input tokens cover material sent to the model. Cached-input pricing applies to input billed as cached, while output tokens cover generated tokens billed in that category. These are separate components of a bill, so a 60% reduction in one component does not mean a 60% reduction overall.
For Astra, the arithmetic is direct:
- Input falls from $10 to $8: 20% less.
- Output falls from $50 to $40: 20% less.
- Cached input falls from $1 to $0.40: 60% less.
Sol keeps its predecessor’s rates in all three categories. Comparing the two new models, Astra costs exactly four times as much per token in each category at these rates.
Heavy Caching Changes the Blended Saving
Consider an illustrative workload spread across multiple short-context requests, with these aggregate token counts:
- 100,000 uncached input tokens
- 900,000 cached-input tokens
- 100,000 output tokens
Using the published rates, the token charges would be:
| Model | Uncached input | Cached input | Output | Total |
|---|---|---|---|---|
| GPT-6 Astra | $1.00 | $0.90 | $5.00 | $6.90 |
| GPT-6.2 Astra | $0.80 | $0.36 | $4.00 | $5.16 |
| GPT-6.2 Sol | $0.20 | $0.09 | $1.00 | $1.29 |
That makes GPT-6.2 Astra about 25.2% cheaper than GPT-6 Astra for this example. Although 90% of the input tokens are cached, output still dominates the bill.
Looking only at input, the reduction is approximately 38.9%, from $1.90 to $1.16. That is the useful distinction: a high cached-token share and a high cached-input spending share are different things.
This calculation holds token counts constant and excludes cache writes, tools and other charges. Actual model behavior can change the totals. The Astra documentation also lists separate cache-write prices of $10 per million tokens for 30-minute retention and $20 for 12-hour retention at short-context Standard rates. Cheap cached reads therefore need to be considered alongside the cost of populating the cache.
Long Context Raises Prices, but Preserves the Discounts
The Astra model documentation specifies a consequential threshold: requests with more than 272,000 input tokens receive higher rates for the full request.
For GPT-6.2 Astra, input and cache rates double, while output costs 1.5 times the short-context rate. That produces these Standard prices:
| Model and context tier | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra, long context | $16.00 | $0.80 | $60.00 |
| GPT-6 Astra, long context | $20.00 | $2.00 | $75.00 |
| GPT-6.2 Sol, long context | $4.00 | $0.20 | $15.00 |
All figures are per million tokens, as listed on OpenAI’s pricing page.
Astra’s relative discounts do not shrink at long context. GPT-6 Astra also has higher long-context rates, so the like-for-like reductions remain 20% for input and output, and 60% for cached input.
The distinction matters for budgeting. A long-context GPT-6.2 Astra request costs more per token than a short-context request, but it retains the same percentage savings against the equivalent GPT-6 Astra tier. Comparing a new model’s long-context price with an old model’s short-context price would mix two different billing conditions.
I’d flag this threshold during migration planning: crossing it changes rates for the full request, not merely the tokens beyond 272,000.
Astra’s Model Card Adds Important Developer Constraints
Astra’s API documentation describes a large-context reasoning model with these limits and capabilities:
- Context window: 1,050,000 tokens.
- Maximum input: 922,000 tokens.
- Maximum output: 128,000 tokens.
- Modalities: Text and image input, with text output; audio and video are unsupported.
- Endpoints: Responses and Chat Completions are supported.
Frequently Asked Questions
3 questions
1What Are the GPT-6.2 API Model IDs?
Use gpt-6.2-astra for Astra and gpt-6.2-sol for Sol.