DEV Community

ModelFild
ModelFild

Posted on

GPT-6.2 on ModelFold: which of the four models to call

OpenAI GPT-6.2 is available on ModelFold today in four models. This note sticks to what the model pages list, and to the calls we have already made.

What GPT-6.2 is

  • GPT-6.2 Sol, model ID gpt-6-2-sol. An upgrade of GPT-6.1 Sol, positioned under the flagship GPT-6.2 Astra, for everyday general work.
  • GPT-6.2 Sol Pro, model ID gpt-6-2-sol-pro. The same model as Sol, served in pro reasoning mode.
  • GPT-6.2 Astra, model ID gpt-6-2-astra. The flagship of the GPT-6.2 series, for the hardest engineering, professional, and research tasks.
  • GPT-6.2 Astra Pro, model ID gpt-6-2-astra-pro. The same model as Astra, served in pro reasoning mode.

Shared capabilities: a 1.05 million token context, image and file input, tool calling, and five reasoning-effort levels (low, medium, high, xhigh, max). The endpoint is the OpenAI-compatible POST /v1/chat/completions. One ModelFold API key calls all four, and the other models on the site.

Pages:

How to choose

Use Sol for everyday questions, a code edit, or summarizing a document.

Use Astra when the task is clearly harder: engineering work that spans many files, professional judgment, or a research-style problem that needs a long chain of reasoning.

Pro is not a different base model. Sol Pro is Sol in pro reasoning mode, and Astra Pro is Astra in pro reasoning mode. Pro spends more reasoning tokens to buy a higher-quality answer. If Sol feels too shallow, try Sol Pro. If you want Astra to go further, try Astra Pro.

Reasoning effort and Pro are separate. Effort is a per-request setting. Pro is a different model ID.

Reasoning effort

Thinking stays on for these four models. You do not turn it off. Set reasoning_effort to one of:

  • low: less thinking, for simple questions.
  • medium: a reasonable default for everyday calls.
  • high: questions that need several steps.
  • xhigh and max: when you want the model to spend the full thinking budget.

Higher levels usually consume more reasoning tokens. Those tokens are billed as output, so a higher effort and a Pro model both cost more. There is no published fixed multiplier. The bill is the usage on that request.

What we have already called successfully: normal chat, streaming, tool calling, and all five effort levels, on all four models. No benchmarks, and no claims about behavior we have not tried.

Price

Pay as you go. No monthly fee. Failed calls are not charged. Prices below are what the model pages showed on 17 November 2026. Use the live page if they change.

  • Sol and Sol Pro: 400 credits per million input tokens (about USD 2), 2000 credits per million output tokens (about USD 10).
  • Astra and Astra Pro: 1600 credits per million input tokens (about USD 8), 8000 credits per million output tokens (about USD 40).

Pro uses the same per-token rates as the matching standard model. It costs more in total because it spends more reasoning tokens, not because the rate is higher.

Python and curl

Create an API key in the ModelFold console and put it in MODELFOLD_API_KEY. The samples use gpt-6-2-sol. To switch models, change only that field.

curl https://modelfold.ai/v1/chat/completions \
  -H "Authorization: Bearer $MODELFOLD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-2-sol","reasoning_effort":"medium","messages":[{"role":"user","content":"Explain idempotency in three sentences."}]}'
Enter fullscreen mode Exit fullscreen mode

For a streaming response, add "stream": true to the same JSON.

from openai import OpenAI

client = OpenAI(base_url="https://modelfold.ai/v1", api_key="YOUR_MODELFOLD_API_KEY")
resp = client.chat.completions.create(
    model="gpt-6-2-sol",
    messages=[{"role": "user", "content": "Explain idempotency in three sentences."}],
    extra_body={"reasoning_effort": "medium"},
)
print(resp.choices[0].message.content)
Enter fullscreen mode Exit fullscreen mode

Tool calling uses the standard tools field on the chat completions API. The model returns the function name and arguments. Your code runs the function. Tool calls succeeded on all four models.

Claude Code and Codex

Do not guess the setup. Follow:

https://modelfold.ai/docs/ai-agent/overview?utm_source=devto

The documented order is: create a key, then export MODELFOLD_API_KEY=your_key. If Node.js is already installed, you can add the skill and let Claude Code or Codex apply the docs:

npx skills add https://modelfold.ai -g -a claude-code -a codex -a grok -y
Enter fullscreen mode Exit fullscreen mode

Claude Code and Codex speak different protocols. Take model IDs from the list endpoint named in that doc. If you only need GPT-6.2 inside your own program, use /v1/chat/completions and set model to gpt-6-2-sol.

Top comments (0)