Requesty
Back|NOV '26AI MODELS / MODEL COMPARISONS
7 MIN READ|

GPT-6.2 Sol and Astra: pricing, benchmarks and reactions

Last updated

In November 2026, GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens at standard API rates, while GPT-6.2 Sol costs $2 and $10. OpenAI announced both models on November 17. Astra's input and output rates are 20% below GPT-6 Astra's, while Sol keeps GPT-6.1 Sol's rates. Independent tests show gains on general intelligence but flat or lower coding scores, and early testers are split.

What changed in the launch?

OpenAI released GPT-6.2 Astra and Sol in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. The API model IDs are gpt-6.2-astra and gpt-6.2-sol, and Pro 500 subscribers can use Ultrafast with either model.

You can also use both models through GitHub Copilot and Amazon Bedrock.

For Sol, the immediate predecessor is GPT-6.1 Sol. For Astra, it is GPT-6 Astra. Our earlier coverage has the GPT-6 Sol launch baseline and the GPT-6 Astra benchmark comparison.

GPT-6.2 pricing against the previous models

The table uses November 2026 standard API prices per million tokens. Cached input is reusable prompt text billed at a lower rate when a cache hit occurs.

ModelInputCached inputOutput
GPT-6.2 Astra$8$0.40$40
GPT-6 Astra$10$1.00$50
GPT-6.2 Sol$2$0.10$10
GPT-6.1 Sol$2$0.10$10
GPT-6 Sol$2$0.20$10

Source: OpenAI API pricing, November 2026.

Astra's input and output rates fell 20%, while its cached input rate fell 60%. Sol's standard input and output rates are unchanged from both GPT-6.1 Sol and GPT-6 Sol, although cached input is half the GPT-6 Sol price.

At the November 2026 standard rates, Astra costs four times as much as Sol for the same number of input or output tokens. That gap can narrow if a model uses fewer output tokens to complete a task: price per token is only part of the bill.

What does a workload cost?

Take 1,000 requests, each using 10,000 uncached input tokens and produces 2,000 output tokens, all at standard rates. Across the workload, that gives 10 million input tokens and 2 million output tokens.

ModelInput cost for 10 million tokensOutput cost for 2 million tokensTotal
GPT-6.2 Astra$80$80$160
GPT-6 Astra$100$100$200
GPT-6.2 Sol$20$20$40

At the standard prices above, GPT-6.2 Astra saves $40 on this workload against GPT-6 Astra, assuming both use the same number of tokens. The example excludes cached input and gateway fees.

The long-context bill can look different. For requests above 272,000 input tokens, long-context rates double the input and cache prices and multiply the output price by 1.5 for the whole request. For GPT-6.2 Astra, that means $16 per million input tokens and $60 per million output tokens.

Service tiers change the bill too: Batch is half the standard rate, while Fast and Ultrafast cost more.

What do the benchmarks say?

OpenAI's launch post reports gains across several evaluations. The figures below are vendor-reported, and GDPval v2 is OpenAI's internal benchmark.

EvaluationOpenAI-reported result
Agents' Last ExamGPT-6.2 Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost.
Terminal-Bench-Science 0.1GPT-6.2 Astra scores 9 points above GPT-6 Astra at less than a third of the cost.
GDPval v2GPT-6.2 Astra gains 6.2 points over GPT-6 Astra with 22% fewer output tokens. GPT-6.2 Sol gains 5.3 points over GPT-6.1 Sol.
FrontierCode 2.0GPT-6.2 Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost, with both at maximum effort.

Task cost can fall by more than the price per token if a model needs fewer tokens to finish. That is how OpenAI can report a larger cost reduction on a particular evaluation than Astra's 20% reduction in standard input and output prices.

Independent tests give us another view of the tradeoff. On the Artificial Analysis Intelligence Index, Astra gained 5 points over GPT-6 Astra and Sol gained 2 points over GPT-6.1 Sol.

ModelNovember 2026 Intelligence Index scoreSource
Claude Fable 5.265Artificial Analysis
GPT-6.2 Astra62Artificial Analysis
Gemini 4 Carbon60Artificial Analysis
Claude Opus 5.559Artificial Analysis
GPT-6.2 Sol57Artificial Analysis

The coding results are less uniform. For its Coding Agent Index, Artificial Analysis tested with the Codex harness at maximum effort: GPT-6.2 Astra was flat against GPT-6 Astra, while Sol lost 3 points against GPT-6.1 Sol. A 14-point drop on SWE-Atlas-QnA-AA drove Sol's decline.

Artificial Analysis also measured hallucination rates on its AA-Omniscience test at maximum effort: 59% for GPT-6.2 Astra and 55% for GPT-6.2 Sol. Astra's result rose from 50% for GPT-6 Astra. These rates describe that test, not error rates in production.

What did early testers say?

Theo said he would not choose GPT-6.2 Astra over Opus 5.5 for his development work. On his TypeScript compiler rewrite in Rust, he said Astra ran for seven days without landing an improvement and made the compiler 56% slower. He also described long tasks where the model blocked itself or went in circles.

His account included successes too. Theo said Astra built a reverse proxy he expects to save him $25k per month, and that it caught code review issues other models missed.

Ethan Mollick found Astra uneven and harder to steer than other models, although it delivered well on a space elevator project built from scratch in Blender and Unity. In another game test, he said it built a working game and strong 3D models, but pursued ideas without asking and left out a shooting mechanism until he pointed it out.

These are early-access anecdotes from a handful of testers. Test on your own tasks before drawing conclusions from them.

What should API users test before switching?

Start with a sample of your own requests. Compare cost per completed task alongside the output, and include the long prompts and coding tasks that are expensive or consequential for your application.

Caching changes the comparison too. OpenAI's prompt caching guide says cache reads on GPT-6.2 Astra and Sol cost 0.05 times the ordinary input rate, so keep reusable prompt prefixes unchanged.

Before moving a workload to GPT-6.2

0 of 5 done

Our earlier guide to GPT-6 Astra access and production readiness covers the previous model's access and safety context.

Which one should you test first?

GPT-6.2 Sol is the lower-cost starting point for a workload comparison. Astra is worth testing on tasks where the added spend could improve the result. Run both against your own prompts, then compare completed-task cost and output quality before changing the default.

We provide one OpenAI-compatible API in front of models from multiple providers, including GPT-6.2 Sol and GPT-6.2 Astra, with routing, fallback policies and cost analytics. If you use a gateway for the comparison, include its fees in the calculation: Requesty charges 5% on top of model costs.

Frequently asked questions
How much do GPT-6.2 Sol and Astra cost?
In November 2026, GPT-6.2 Sol costs $2 per million input tokens and $10 per million output tokens at standard rates. GPT-6.2 Astra costs $8 and $40 respectively.
Is GPT-6.2 Astra cheaper than GPT-6 Astra?
At November 2026 standard rates, GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra's $10 and $50 rates.
How does GPT-6.2 Sol compare with GPT-6 Sol on price?
At November 2026 standard rates, GPT-6.2 Sol and GPT-6 Sol both cost $2 per million input tokens and $10 per million output tokens. GPT-6.2 Sol's cached input costs $0.10 per million tokens, compared with $0.20 for GPT-6 Sol.
How do GPT-6.2 Astra and Sol score on independent benchmarks?
In November 2026, Artificial Analysis scored GPT-6.2 Astra at 62 and GPT-6.2 Sol at 57 on its Intelligence Index. Claude Fable 5.2 scored 65.
What are the API model IDs for GPT-6.2 Sol and Astra?
OpenAI's API model IDs are gpt-6.2-sol and gpt-6.2-astra. OpenAI announced both models on November 17, 2026.
Related reading

Start building with Requesty

One line of code. 600+ models. Full control.

Speak to founders