
The GPT-6.2 API is live on PoYo with two models: gpt-6.2-astra for the hardest work and gpt-6.2-sol for everyday coding and agent tasks. Both run on the OpenAI-compatible Chat Completions and Responses endpoints with one PoYo key, at 20% below OpenAI's standard token rates. You can try each one on the GPT-6.2 Astra model page and the GPT-6.2 Sol model page before you write code.
This overview covers what OpenAI shipped, the limits you have to design around, the price on PoYo and at OpenAI, and which requests belong on which model.
What is GPT-6.2?
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. OpenAI's API changelog describes the split in one line: "Use GPT-6.2 Astra for demanding reasoning and coding tasks, or GPT-6.2 Sol for everyday coding and agent workflows." Astra succeeds GPT-6 Astra and Sol succeeds GPT-6.1 Sol.
| Date | What happened |
|---|---|
| November 17, 2026 | OpenAI releases GPT-6.2 Astra and GPT-6.2 Sol in its API (Chat Completions, Responses, Batch) |
| November 18, 2026 | Both callable on PoYo as gpt-6.2-astra and gpt-6.2-sol |
Both are reasoning models. They take text and images, return text, and share the same context and output limits. The differences that matter to an integration are price per token (Astra costs four times as much as Sol) and how much capability you need per request.
GPT-6.2 API at a glance
| Item | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| API model ID | gpt-6.2-astra | gpt-6.2-sol |
| PoYo model page | /models/gpt-6-2-astra | /models/gpt-6-2-sol |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Max input | 922,000 tokens | 922,000 tokens |
| Max output | 128,000 tokens | 128,000 tokens |
| Input / output | Text and image in, text out | Text and image in, text out |
| Knowledge cutoff | May 18, 2026 | June 1, 2026 |
| Reasoning effort | none is rejected; OpenAI recommends starting at medium | Same |
| Tool calling | Responses only | Responses only |
| PoYo formats | Chat Completions, Responses | Chat Completions, Responses |
Limits and cutoffs come from OpenAI's GPT-6.2 Astra and GPT-6.2 Sol model pages. The 1.05M window is total capacity: a request can carry up to 922K input tokens and must leave room for the answer, which includes hidden reasoning tokens.
Endpoints and authentication
Both formats use Authorization: Bearer $POYO_API_KEY against https://api.poyo.ai. Responses come back in each format's native shape, without the code / data wrapper PoYo uses for media tasks.
| Format | Endpoint | Streaming | Use it for |
|---|---|---|---|
| Responses | POST https://api.poyo.ai/v1/responses | "stream": true (SSE) | Function tools, image input, reasoning.effort, new builds |
| Chat Completions | POST https://api.poyo.ai/v1/chat/completions | "stream": true (SSE) | Existing OpenAI SDK code with plain text messages |
OpenAI's GPT-6.2 guide is explicit: "GPT-6.2 Astra and GPT-6.2 Sol support Chat Completions, but tool calling requires Responses." PoYo's Responses reference documents tools, tool_choice, reasoning, previous_response_id, structured input with input_image blocks, and streaming. PoYo's Chat Completions reference documents string content only, so send images through Responses. Request shapes are in the OpenAI models docs.
Three changes in OpenAI's GPT-6.2 guide affect existing code:
- Omit sampling fields. For Chat Completions, "omit
temperature,top_p,logprobs, andtop_logprobs." For Responses, omittemperature,top_p, andtop_logprobs. - Reasoning effort.
noneis rejected. Start withmedium; if you usednoneorminimal, uselow. - Instructions placement. "Models starting with GPT-6.2 append top-level
instructionsafterinput." Put stable instructions in adevelopermessage at the start ofinputso they stay in the cacheable prefix.
GPT-6.2 API pricing: PoYo vs OpenAI
PoYo bills per token in credits (1 credit = $0.005). Prices per 1M tokens:
| Meter | Astra on PoYo | Astra at OpenAI | Sol on PoYo | Sol at OpenAI |
|---|---|---|---|---|
| Input | $6.40 (1,280 credits) | $8.00 | $1.60 (320 credits) | $2.00 |
| Output, incl. reasoning | $32.00 (6,400 credits) | $40.00 | $8.00 (1,600 credits) | $10.00 |
| Cache read | $0.32 (64 credits) | $0.40 | $0.08 (16 credits) | $0.10 |
| Cache write | $8.00 (1,600 credits) | $10.00 | $2.00 (400 credits) | $2.50 |
OpenAI's prices are its standard short-context rates. OpenAI bills prompts above 272K input tokens at 2× the input and cache rates and 1.5× the output rate for the whole request, and its reasoning guide says reasoning tokens are billed as output. PoYo prices are as listed on November 18, 2026; the model pages are the current source.
A 150K-token code review with a 12K-token answer costs 0.15 × $1.60 + 0.012 × $8.00 = $0.336 (67.2 credits) on Sol and 0.15 × $6.40 + 0.012 × $32.00 = $1.344 (268.8 credits) on Astra. More examples, including cached agent sessions, are in GPT-6.2 API pricing.
Which requests to route to each GPT-6.2 API model
OpenAI's GPT-6.2 guide says: use GPT-6.2 Astra "for the most demanding tasks" or GPT-6.2 Sol "for complex work at lower cost." OpenAI's GPT-6.2 system card gives a few side-by-side numbers:
| Evaluation (system card) | Astra | Sol | Routing signal |
|---|---|---|---|
| GPQA Diamond (no tools) | 97.0% | 96.0% | Hard science Q&A: Sol is close |
| MLE-Bench Revised (mean reference percentile) | 96.06 | 91.38 | ML engineering: Astra ahead |
| Sandbox Bench (targets solved) | 36.4% | 25.0% | Long multistep security tasks: Astra ahead |
| SEC-Bench Pro (vulnerability reproduction) | 86.75% | 88.34% | Sol slightly ahead |
These come from a safety evaluation document, not a product benchmark page, and the system card does not state the reasoning effort for most of them. Treat them as a hint and test your own prompts. A practical default: send everything to gpt-6.2-sol, and escalate to gpt-6.2-astra when a task fails your checks or is known to be hard. The full routing setup is in GPT-6.2 Sol API vs GPT-6.2 Astra API.
GPT-6.2 API quick start
curl https://api.poyo.ai/v1/responses \
-H "Authorization: Bearer $POYO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6.2-sol",
"input": [
{"role": "developer", "content": "You are a senior engineer. Answer with a numbered plan."},
{"role": "user", "content": "Plan a migration of a Flask service from SQLAlchemy 1.4 to 2.0."}
],
"reasoning": {"effort": "medium"},
"max_output_tokens": 8000
}'
The answer is in the output[].content[] blocks with type: "output_text", and token counts are in usage. Swap the model for gpt-6.2-astra to run the same request on Astra. The OpenAI SDK, streaming, image input, and Chat Completions are in the GPT-6.2 API guide. For tools and long-running agents, see GPT-6.2 API for coding agents. Moving from an older GPT-6 model? Read GPT-6.2 API vs GPT-6.1 Sol API.
Frequently asked questions
What are the GPT-6.2 API model IDs?
Use gpt-6.2-astra and gpt-6.2-sol in the model field of /v1/responses or /v1/chat/completions. The PoYo model pages are at /models/gpt-6-2-astra and /models/gpt-6-2-sol.
How much does the GPT-6.2 API cost?
On PoYo, per 1M tokens: Astra $6.40 input, $32 output, $0.32 cache read, $8 cache write; Sol $1.60, $8, $0.08, $2. OpenAI's standard rates are 25% higher: $8 / $40 / $0.40 / $10 and $2 / $10 / $0.10 / $2.50.
Is the GPT-6.2 API compatible with the OpenAI SDK?
Yes. Set base_url="https://api.poyo.ai/v1" and use your PoYo key. Call client.responses.create for tools and images, or client.chat.completions.create for text chat.
What is the GPT-6.2 context window and max output?
1,050,000 tokens of context, with up to 922,000 input tokens and 128,000 output tokens, for both Astra and Sol.
Does the GPT-6.2 API support function calling?
Yes, through the Responses API. OpenAI states that tool calling for GPT-6.2 requires Responses; Chat Completions works for requests without tools.
Should I use GPT-6.2 Astra or GPT-6.2 Sol?
Start with Sol, which costs a quarter of Astra per token. Move a request type to Astra when Sol's output fails your tests, or when the task is your hardest work: complex code changes, long professional deliverables, or research.