OpenAI's top model
OpenAI's own charts put it ahead of GPT-6 Astra and GPT-6.2 Sol on agent, coding and science benchmarks.
GPT-6.2 Astra API: OpenAI's most capable model for the hardest projects, from start to finish.
gpt-6.2-astraUpdated from openai import OpenAIclient = OpenAI( base_url="https://api.zurelay.com/v1", api_key="YOUR_ZURELAY_KEY",)stream = client.chat.completions.create( model="gpt-6.2-astra", messages=[ {"role": "user", "content": "Explain quicksort briefly."}, ], stream=True,)for chunk in stream: if chunk.choices: print(chunk.choices[0].delta.content or "", end="")Base URL https://api.zurelay.com/v1. Works with any OpenAI SDK. Chat API docs
Pricing
Pay as you go from prepaid credit, with no subscription. Top up from $10. Requests that fail are never billed.
| Rate | Zurelay | OpenAI list | You save |
|---|---|---|---|
Input per 1M tokens | $2.00 | 75%off | |
Output per 1M tokens | $10.00 | 75%off | |
Input, long prompts per 1M tokens | $4.00 | 75%off | |
Output, long prompts per 1M tokens | $15.00 | 75%off | |
Cached input per 1M tokens | $0.10 | 75%off |
Prompts over 272K tokens cost 2 times the input rates, cached input included, and 1.5 times the output rate, the way OpenAI bills them. Streaming, tool calls and structured outputs cost nothing extra.
Every GPT model and priceMove the sliders to your monthly usage.
$7,200 a year
Cost
Typical workloads at Zurelay's live prices next to the list price, and what other places charge for the same model.
50,000 requests a month, 1,500 tokens in and 300 out, 30% of the input cached. $0.00515 a request.
20,000 requests a month, 8,000 tokens in and 500 out, 20% of the input cached. $0.018 a request.
5,000 requests a month, 40,000 tokens in and 1,500 out, 85% of the input cached. $0.0304 a request.
1,000,000 requests a month, 400 tokens in and 5 out. $0.00085 a request.
Prices per 1M tokens
| Seller | Input | Output | vs OpenAI |
|---|---|---|---|
OpenAI List price, direct | $8.00 | $40.00 | — |
List price, plus 5.5% on card top-ups | $8.00 | $40.00 | +5.5% |
Zurelay No fee to add credit | $2.00 | $10.00 | 75%off |
OpenRouter's fee as its pricing page listed it on November 15, 2026. Zurelay's prices are live. Some labs also discount batch or off-peak requests.
Overview
GPT-6.2 Astra is OpenAI's most capable model, released on November 17, 2026, for software, documents, spreadsheets, presentations and research. It replaces GPT-6 Astra at OpenAI's top tier, at a list price 20% lower. The GPT-6.2 Astra API on Zurelay serves it through one OpenAI-compatible endpoint, below OpenAI's list price.
GPT-6.2 Astra is the successor to GPT-6 Astra, released by OpenAI on November 17, 2026, together with GPT-6.2 Sol. OpenAI calls it its most capable model for your hardest projects, from start to finish: it builds and revises software, uses a computer, and works with documents, spreadsheets, presentations and scientific research. The GPT-6.2 Astra API on Zurelay uses the model ID gpt-6.2-astra on an OpenAI-compatible Chat Completions endpoint.
The limits match the rest of GPT-6: text and image input, text output, a 1,050,000-token context window, up to 922,000 input tokens and up to 128,000 output tokens. The knowledge cutoff is May 18, 2026. Reasoning effort accepts low, medium (the default), high, xhigh and max. OpenAI's API rejects none for this model, so Zurelay runs a request for none at low.
OpenAI reports gains over GPT-6 Astra at maximum reasoning effort: 77.6 against 73.1 on Agents' Last Exam v1.1, 77.1 against 68.1 on Terminal-Bench-Science 0.1, 44.2 against 42.8 on FrontierCode 2.0 and 86.1 against 79.8 on LifeVal. Its system card puts it at 97.0% on GPQA Diamond without tools. OpenAI also says it scores 6.2 points higher on GDPval v2 with 22% fewer output tokens. These are OpenAI's own evaluations.
OpenAI lists it at $8 per 1M input tokens and $40 per 1M output tokens, 20% below GPT-6 Astra, with cached input at 5% of the input rate. OpenAI rates GPT-6.2 at its Critical level for cybersecurity capability and has tightened its default limits on higher-risk security help, so expect more refusals on offensive security requests than with earlier models.
Strengths
OpenAI's own charts put it ahead of GPT-6 Astra and GPT-6.2 Sol on agent, coding and science benchmarks.
OpenAI says it scores higher than GPT-6 Astra on GDPval v2 while writing 22% fewer output tokens.
OpenAI says it is better at changing software that already exists, not just writing new code.
OpenAI cut Astra's list price from $10 and $50 to $8 and $40, and Zurelay charges less than that.
Use cases
Plan and make changes across an existing project, review pull requests and fix bugs that need the whole picture.
Draft reports, analyse spreadsheets and outline presentations close to professional templates.
Long, careful reasoning over papers, data and technical material, with up to 922,000 tokens of input.
Multi-step tool-calling work where getting it right the first time matters more than the price per token.
Get started
No waitlist and no new SDK. If your code already talks to OpenAI, it already talks to Zurelay.
Sign up, add credit and create an API key. Set a monthly budget or a rate limit per key if you like.
Change the base URL. Everything else in your code stays the same.
https://api.zurelay.com/v1Send chat completions as usual. Streaming, tool calls and usage reporting work as you expect.
gpt-6.2-astraWorks with the tools you already use
Compare
$0.50 / $2.50 per 1M tokens
Pick it for everyday coding and agents: OpenAI says it comes within half a point of Astra on FrontierCode 2.0 at a quarter of Astra's list price.
GPT-6.2 Sol API$2.50 / $12.50 per 1M tokens
Pick it only to keep behavior you have already tested: 6.2 Astra costs less and scores higher.
GPT-6 Astra API$0.25 / $1.25 per 1M tokens
Pick it for fast, high-volume agent and vision work at one-eighth of 6.2 Astra's list price per token.
GPT-6.1 Terra APIOn Zurelay, GPT-6.2 Astra costs $2.00 per 1M input tokens and $10.00 per 1M output tokens. OpenAI's list price is $8.00 input and $40.00 output, so you save 75%. Cached input tokens bill at $0.10 per 1M, and reasoning tokens count as output tokens. Prompts over 272K tokens bill at $4.00 input and $15.00 output for the whole request, as at OpenAI.
Zurelay serves gpt-6.2-astra at 75% below OpenAI's list price, with the same model ID and request format. For everyday coding and agent work, GPT-6.2 Sol costs a quarter as much per token.
Install the official openai package and create a client with base_url set to https://api.zurelay.com/v1 and your Zurelay API key. Then send a Chat Completions request with model set to "gpt-6.2-astra". Set reasoning_effort to low, medium, high, xhigh or max, and cap output with max_completion_tokens.
GPT-6.2 Astra has a 1,050,000-token context window. One request can include up to 922,000 input tokens, and the model can write up to 128,000 output tokens. Input can be text, images or PDF files; output is text.
OpenAI's list price per token is 20% lower, $8 and $40 against $10 and $50, and cached input is $0.40 against $1. At maximum effort OpenAI reports 77.6 against 73.1 on Agents' Last Exam v1.1 and 77.1 against 68.1 on Terminal-Bench-Science 0.1. The knowledge cutoff moved to May 18, 2026. Both stay on Zurelay, so code tuned on GPT-6 Astra keeps working while you test 6.2.
OpenAI lists Sol at a quarter of Astra's price per token and calls it the go-to model for everyday work. On FrontierCode 2.0 at maximum effort Sol scored 43.8 against Astra's 44.2; Astra leads more on Terminal-Bench-Science (77.1 against 68.0) and LifeVal (86.1 against 83.8). Run both on a sample of your real prompts and pay for Astra only where it wins.
OpenAI rates GPT-6.2 at its Critical level for cybersecurity capability and tightened its default limits on higher-risk security help at launch. The model can still explain concepts and help with defensive work, but it refuses more offensive requests than earlier GPT models. Those limits come from OpenAI's model, not from Zurelay.
Yes. Set stream: true to receive tokens as they are generated, and add stream_options.include_usage to get token counts in the final chunk. You can give each Zurelay API key its own requests-per-minute limit and monthly budget in the dashboard. If an upstream call errors or times out, Zurelay retries it on another route before returning an error.
Yes. Requests to gpt-6.2-astra run on OpenAI's GPT-6.2 Astra. Zurelay doesn't swap in a smaller or cheaper model. Only smart routing, which you turn on yourself, can answer with a close alternative when the model is down, and the x-zurelay-model header names the model that answered. As with any call to OpenAI, two responses to the same prompt can differ because of sampling. Zurelay is independent and is not affiliated with or endorsed by OpenAI.
Create a key in seconds, point the SDK you already use at Zurelay, and every request costs up to 90% less from the first token.