Summary:

Two new models bring different price points. Preserving conversation state and testing the full cost of a completed task matter more than changing a model name.

OpenAI has added GPT-6.2 Astra and GPT-6.2 Sol to its API lineup in a November 17 changelog entry. Both work through Responses and Chat Completions, but applications that call tools need the Responses API. The practical migration question is not just which model to choose: it is whether an existing application preserves the conversation history the new models rely on. OpenAI API changelog.

Astra and Sol: the documented differences

OpenAI positions Astra for demanding reasoning and coding, and Sol for everyday coding and agent workflows. Both accept text and images and return text. Their model pages list a 1,050,000-token context window, a maximum of 922,000 input tokens and up to 128,000 output tokens. Neither accepts reasoning.effort: "none".

Standard price per 1M tokensGPT-6.2 AstraGPT-6.2 Sol
Input$8$2
Cached input$0.40$0.10
Output$40$10

These are USD Standard rates for requests with up to 272,000 input tokens. Bold identifies the lower price, not a quality winner. At the same token counts, Sol's listed input and output rates are one-quarter of Astra's. Actual task cost also depends on reasoning, repeated work and tool charges. Sources: Astra model documentation and Sol model documentation.

The migration detail to check first

OpenAI's GPT-6.2 guide says the API drops reasoning items when it detects edits to earlier conversation history, a model downgrade, a changed end user or a changed organization. Losing that state can reduce answer quality and make the model repeat work using additional tokens.

That makes conversation handling an important part of a migration. An application that silently rewrites earlier instructions or tool messages should review those changes rather than assume that swapping the model name is the whole upgrade. OpenAI recommends keeping the existing prefix unchanged when adjusting settings between responses. Its guide also describes Astra as more concise by default, so instructions that repeatedly demand shorter answers may need retuning. GPT-6.2 migration and prompting guide.

Check the pricing boundary, not only the headline rate

For requests above 272,000 input tokens, both model pages specify twice the input and cache rates and 1.5 times the output rate for the entire request. A large context window therefore does not mean a flat price across its full range.

The pricing page also lists discounted background requests: eligible two-hour completion windows are 30% below Standard model-token rates, and eight-hour windows are 40% below. These are alternatives for work that can wait, not promises of faster interactive responses. Current pricing and processing options.

A useful first comparison

Run the same representative tasks through both candidates with fixed inputs, equivalent tool access and clear completion criteria. Record successful outcomes, elapsed time, total billed tokens and retries. Include a long conversation and a tool-heavy task so the test exercises state handling as well as the final answer.

That separates a cheaper token rate from a cheaper completed job, a distinction that also matters in request-share versus spending comparisons. Keep a working fallback while testing the new model routes.

Sources and image credit

This report uses OpenAI's November 17 API changelog, model documentation, migration guide and pricing page linked above. Prices and access details were checked November 17, 2026. The cover is an original PromptsZone price chart, not a benchmark scorecard. The official OpenAI provider wordmark is used under OpenAI's brand guidelines; it belongs to OpenAI and does not imply endorsement.