Skip to main content
OpenAI released GPT-6.2 on November 17, 2026: Astra for the hardest work and Sol for everyday use. Both run on any default-group key. Astra costs $8 input and $40 output per 1M tokens, 20% less than GPT-6 Astra; Sol costs $2 and $10, the same as GPT-6.1 Sol.
  • Published: November 18, 2026
  • Last verified: November 18, 2026
  • Status: Active; available in the default token group

Key facts

Astra or Sol?

  • gpt-6.2-sol: OpenAI’s everyday model, balancing capability and cost. A good default for chat, writing, analysis, and most coding. Knowledge cutoff: June 1, 2026.
  • gpt-6.2-astra: OpenAI’s most capable model, for complex reasoning and hard coding problems. It costs four times as much as Sol. Knowledge cutoff: May 18, 2026.
  • High-volume, high-concurrency work: OpenAI still points to GPT-6 Luna, so keep using gpt-6-luna.

Pricing

These rates come from LaoZhang API’s public pricing as of November 18, 2026, in US dollars per 1M tokens. They match the Standard rates on OpenAI’s pricing page.
  • Requests with up to 272K input tokens use the standard rates. Above 272K input tokens, the whole request is billed at the long-context rates.
  • Token usage and the charge for each request are in call logs.
  • For current rates, see the model and pricing catalog.

How GPT-6.2 compares with GPT-6.1 Sol and GPT-6 Astra

GPT-6.2 Sol vs GPT-6.1 Sol

  • Every rate is identical. To switch, change model; your key and endpoint stay the same.
  • Reasoning effort levels and tool calling in Chat Completions also work the same way, so existing GPT-6.1 Sol requests need no parameter changes.

GPT-6.2 Astra vs GPT-6 Astra

  • Moving to GPT-6.2 Astra cuts input, output, and cache-write rates by 20%, and cache reads by 60%.
  • If you send "reasoning_effort": "max" through Chat Completions today, move those requests to the Responses API when you switch to gpt-6.2-astra; see the next section.

Reasoning effort and parameters

Neither model accepts none, and the two endpoints offer different levels. If you need max from Chat Completions, use the Responses API instead. Sending none returns 400 on either endpoint. From the Responses API with gpt-6.2-sol, the error reads:
When you migrate from an older model, also check:
  • Sampling parameters such as temperature and top_p: leave them out of the request, or set them to “not set” in your client.
  • max_tokens: keep it at 128,000 or below; larger values return 400.

Call GPT-6.2

Create an API key in Token management in the default group, with its billing mode set to Usage first (按量优先, recommended) or Usage-based (按量计费). Then export the key in your terminal:
These examples call gpt-6.2-sol through Chat Completions and through Responses; the Responses example raises reasoning effort to high. For Python, run python -m pip install openai first.
  • For the flagship model, change model to gpt-6.2-astra; nothing else changes.
  • In Chat Completions, the reply text is in choices[0].message.content.
  • In Responses, the text is in the output item whose type is message; the Python SDK joins it in response.output_text.
  • For requests with tools, see function calling; both endpoints work.

Who is affected

  • Affected: projects that call text models through LaoZhang API’s OpenAI-compatible endpoint. Any API key in the default group with the Usage first or Usage-based billing mode can call GPT-6.2.
  • Not affected: existing models. Every GPT-6.1 and GPT-6 model remains available at the same price, and current integrations need no changes.

Frequently asked questions

Should I use GPT-6.2 Astra or Sol?

Start with gpt-6.2-sol. It is OpenAI’s everyday model, costs a quarter of Astra’s rates, and handles most chat, writing, and coding work. Try gpt-6.2-astra on the complex reasoning or hard coding tasks where Sol falls short, and compare quality against total cost on your own samples. For cheap bulk work, use gpt-6-luna.

How much does the GPT-6.2 API cost?

On LaoZhang API, gpt-6.2-astra costs $8 input and $40 output per 1M tokens, and gpt-6.2-sol costs $2 and $10, matching OpenAI’s Standard rates as of November 18, 2026. Cache and long-context rates are in Pricing.

GPT-6.2 Sol costs the same as GPT-6.1 Sol. Should I switch?

There’s no rush: prices and request format are the same, and gpt-6.1-sol stays available. To try the new version, change model and compare quality, latency, and token usage on representative requests before you commit.

Why does reasoning_effort “none” return 400?

Neither GPT-6.2 model supports none. If your requests send none or minimal, change them to the lowest level, low.

Can GPT-6.2 call tools in Chat Completions?

Yes. OpenAI says tool calling needs the Responses API, but LaoZhang API’s Chat Completions endpoint accepts requests with tools and returns tool_calls, as it does for GPT-6.1. Clients that only speak Chat Completions can use GPT-6.2 with tools; see function calling for an example.

How are requests over 272K input tokens billed?

The whole request is billed at the long-context rates, not just the tokens above 272K. Input and cache rates double, and output rates rise 1.5×; for gpt-6.2-sol, output goes from $10 to $15 per 1M tokens.