· reAPI Team
GPT-6.2 API Pricing: Astra and Sol Cost Per Task
GPT-6.2 API pricing explained: Astra at $8/$40 and Sol at $2/$10 per million tokens, plus cache, batch, long-context and Ultrafast rates with worked costs.

GPT-6.2 API pricing starts at $8 per million input tokens and $40 per million output tokens for GPT-6.2 Astra, and $2 and $10 for GPT-6.2 Sol, at Standard speed for prompts up to 272K input tokens[1][2]. Astra is 20% cheaper than GPT-6 Astra on both lines. Sol costs the same as GPT-6.1 Sol.
Those two headline numbers decide less of your bill than they seem to. Cached input, cache writes, the 272K long-context boundary, batch, and the Fast and Ultrafast speed tiers each change what a request costs, sometimes by a factor of six. This guide lists every published rate we could trace, then prices three concrete workloads so you can see where the money goes.
GPT-6.2 API pricing at a glance
All figures are US dollars per million tokens at Standard speed, for requests with no more than 272K input tokens[1][5].
| Model | Input | Cached input | Cache write (30 min) | Output |
|---|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $0.40 | $10.00 | $40.00 |
| GPT-6 Astra (previous) | $10.00 | $1.00 | $12.50 | $50.00 |
| GPT-6.2 Sol | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6.1 Sol (previous) | $2.00 | $0.10 | $2.50 | $10.00 |
Two details are easy to miss. Astra's cached input fell further than its headline rate: from $1.00 to $0.40, a 60% cut rather than 20%[3]. And Sol's price did not move at all, so any Sol savings come from the model finishing tasks with fewer tokens, which is OpenAI's claim, not a rate change[6].
Per token, Sol is one quarter of Astra on input, cached input and output[7]. Under GPT-6 Astra the gap was one fifth, so the cheaper Astra narrows the distance between the two models.
Both models share a 1,050,000-token context window, a 128,000-token output limit and a 922,000-token maximum input[4][5].
GPT-6.2 API pricing beyond the headline: cache, batch, long context and speed
Cache writes
Writing to the prompt cache is billed separately from reading it. A 30-minute cache write costs $10.00 per million tokens on Astra and $2.50 on Sol. A 12-hour write on short-context Standard requests costs $20.00 on Astra and $5.00 on Sol[1][5]. On Astra, a 30-minute write costs 25 times a cached read, so caching pays off only when the same prefix is read back several times within the window.
Batch and background processing
Batch requests cost half of Standard on both models: $4 input and $20 output for Astra, $1 and $5 for Sol[1]. OpenAI's changelog for launch day also added background completion windows in public beta, billed 30% below Standard for a two-hour window and 40% below for an eight-hour window[5].
Prompts over 272K input tokens
Once a request's input passes 272K tokens, the input and cache rates double and output is billed at 1.5 times, for the whole request rather than only the tokens beyond the line[1][5]. For Astra that means $16 input, $0.80 cached input and $60 output. Kingy AI's reading of the price list puts Sol at $4 input and $15 output in the same band[4].
Fast and Ultrafast
Fast mode, the tier OpenAI called Priority processing until July 2026, costs twice the Standard rate on both models[1]. Ultrafast costs three times Standard through February 28, 2027 and six times Standard from March 1, 2027, Pacific time[1][5].
| Ultrafast rate | Input | Output |
|---|---|---|
| GPT-6.2 Astra, until Feb 28, 2027 | $24.00 | $120.00 |
| GPT-6.2 Astra, from Mar 1, 2027 | $48.00 | $240.00 |
| GPT-6.2 Sol, until Feb 28, 2027 | $6.00 | $30.00 |
| GPT-6.2 Sol, from Mar 1, 2027 | $12.00 | $60.00 |
Data Studios also reports a 10% uplift for regional processing endpoints used for data residency, applied to models released after March 5, 2026[1].
What a task costs on GPT-6.2 Astra and Sol
The worked examples below apply GPT-6.2 API pricing to three request shapes, using our own arithmetic from the rates above. Token counts are illustrative, chosen to show how each pricing rule behaves; they are not measurements. Cache writes are excluded.
A coding-agent turn with a warm cache
Assume 150,000 input tokens, of which 105,000 hit the cache, and 12,000 output tokens. The request stays under 272K, so Standard rates apply.
| Model | Uncached input (45K) | Cached input (105K) | Output (12K) | Total |
|---|---|---|---|---|
| GPT-6 Astra | $0.450 | $0.105 | $0.600 | $1.155 |
| GPT-6.2 Astra | $0.360 | $0.042 | $0.480 | $0.882 |
| GPT-6.2 Sol | $0.090 | $0.011 | $0.120 | $0.221 |
The saving from GPT-6 Astra to GPT-6.2 Astra is 23.6% on this shape of request, slightly better than the 20% headline because of the deeper cached-input cut. Across 1,000 such turns the bill drops from $1,155 to $882. The same 1,000 turns on Sol cost about $221.
Data Studios priced a heavier agent workload, 2,000 requests a day at 60,000 input tokens with 80% cached and 3,000 output tokens, and arrived at $636.00 a day on GPT-6 Astra, $470.40 on GPT-6.2 Astra (26% lower) and $117.60 on Sol[1]. The more of your input is cached, the larger Astra's saving.
A long-document request over 272K
Assume 400,000 uncached input tokens and 20,000 output tokens. Because input is above 272K, the long-context rates apply to the entire request.
- GPT-6.2 Astra: 400,000 × $16 / 1M = $6.40, plus 20,000 × $60 / 1M = $1.20, for $7.60.
- Priced at Standard rates by mistake, the same request would show $4.00, understating the bill by $3.60.
- GPT-6.2 Sol, using Kingy AI's long-context figures: $1.60 + $0.30 = $1.90.
If you can split a large document into chunks that each stay under 272K input tokens, every chunk is billed at the Standard rate. Whether chunking preserves enough context for your task is a quality question you need to test, but the price difference is large enough to justify the test.
The same request at different speeds
Kingy AI used a simple reference request of 100,000 uncached input tokens and 20,000 output tokens, which costs $1.60 on Astra and $0.40 on Sol at Standard speed[4]. Applying the published multipliers:
| Speed | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| Batch (0.5×) | $0.80 | $0.20 |
| Standard (1×) | $1.60 | $0.40 |
| Fast (2×) | $3.20 | $0.80 |
| Ultrafast until Feb 28, 2027 (3×) | $4.80 | $1.20 |
| Ultrafast from Mar 1, 2027 (6×) | $9.60 | $2.40 |
The spread between the cheapest and most expensive way to run one Astra request is twelvefold. Until March, Ultrafast Sol ($1.20) costs less than Standard Astra ($1.60) on this request; from March 1 it costs more ($2.40). A latency-sensitive feature that tolerates Sol's quality can buy speed more cheaply on Sol than on Astra at any speed above Standard.
How GPT-6.2 prices compare with Claude and Gemini
At list price, GPT-6.2 API pricing for Astra matches Anthropic's Claude Fable 5.2 at $8 input and $40 output per million tokens, and Claude Opus 5.5 sits at $4 and $20[1][4]. Fable's cached read is cheaper than Astra's: $0.20 against $0.40[4]. For cache-heavy agents that difference matters more than the matching headline rates.
Per-token rates are only half of a comparison, because models spend different numbers of tokens on the same task. On Artificial Analysis's Intelligence Index at maximum effort, Kingy AI recorded a weighted cost per task of $4.00 for Astra, $1.01 for Sol and $26.54 for Fable 5.2[4]. The Decoder reports OpenAI's claim that Astra beats Fable 5.2 by 5.5 points on Agents' Last Exam at about 5% of the cost[2]. That is a vendor benchmark, so treat it as a reason to test, not as a result for your workload.
Data Studios lists Gemini 4 Argon at $2 input and $10 output until November 30, rising to $4 and $20 from December 1[1]. If you are comparing Sol with Argon, check which period your estimate falls in.
Cutting a GPT-6.2 bill without changing models
- Raise the cache hit rate. On Astra a cached read is one twentieth of a fresh input token. Keep stable system prompts, tool definitions and reference documents at the start of the prompt so they form a reusable prefix.
- Stay under 272K where the task allows it. Crossing the line raises the rate for the whole request, not just the overflow.
- Move non-urgent work to batch or background windows. Batch halves the bill; the background windows take 30% or 40% off[1][5].
- Reserve Ultrafast for user-facing latency. It triples the price today and will cost six times Standard after February 2027.
- Route by step, not by product. Many pipelines can send planning or hard reasoning steps to Astra and routine steps to Sol, which costs a quarter as much per token[7].
FAQ
How much does GPT-6.2 Astra cost?
$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens at Standard speed for prompts up to 272K input tokens[1].
How much does GPT-6.2 Sol cost?
$2 per million input tokens, $0.10 cached and $10 output, the same as GPT-6.1 Sol[2].
Is GPT-6.2 Astra cheaper than GPT-6 Astra?
Yes. Input and output are 20% lower and cached input is 60% lower[3]. On cache-heavy requests the real saving is a little above 20%, as the worked example shows.
What happens above 272K input tokens?
Input and cache rates double and output costs 1.5 times Standard, applied to the entire request[5].
How much does Ultrafast cost?
Three times Standard through February 28, 2027, then six times Standard from March 1, 2027, Pacific time[1].
Does batch pricing apply to GPT-6.2?
Yes. Batch is half of Standard on both Astra and Sol[1].
What are the GPT-6.2 context and output limits?
A 1,050,000-token context window, up to 922,000 input tokens and up to 128,000 output tokens on both models[5].
Reading GPT-6.2 API pricing for your own workload
Start from your real request shape: how many input tokens, how much of it repeats, how much output, and how fast the answer has to arrive. Those four numbers decide the bill more than the $8 and $2 headlines do. For most cache-friendly agent work, GPT-6.2 Astra now costs a little under four fifths of what GPT-6 Astra did, and Sol does the same token work for a quarter of Astra's price. Keep requests under 272K input tokens when you can, push anything non-interactive to batch, and treat Ultrafast as a feature cost. Check the official price list again before you commit a budget, because GPT-6.2 API pricing for Ultrafast is already scheduled to change on March 1, 2027.
References
- Data Studios. OpenAI launches GPT-6.2 Astra and Sol, cutting Astra API prices 20%. Retrieved November 17, 2026 from datastudios.org
- The Decoder. OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gains. Retrieved November 17, 2026 from the-decoder.com
- Crypto Briefing. OpenAI launches GPT-6.2 Astra and Sol with 20% lower API pricing. Retrieved November 17, 2026 from cryptobriefing.com
- Kingy AI. GPT-6.2 Astra and Sol benchmarks, pricing and reviews. Retrieved November 17, 2026 from kingy.ai
- Munder Difflin. GPT-6.2 Astra and Sol: Price, Specs and How to Use Them in Codex. Retrieved November 17, 2026 from munderdiffl.in
- Thurrott. OpenAI Launches GPT-6.2 Astra and Sol Models in ChatGPT Work and Codex. Retrieved November 17, 2026 from thurrott.com
- AIFix (DEV Community). gpt 6.2 for Automation Builders: Astra vs Sol, Prices and a Simple Routing Setup. Retrieved November 17, 2026 from dev.to