DEV Community

Twinkle Model
Twinkle Model

Posted on Edited on

Breaking: GPT-6.2 Astra and Sol are live on Twinkle Model

OpenAI released gpt-6.2-astra and gpt-6.2-sol on November 17, 2026. Twinkle Model connected both official model IDs on its OpenAI-compatible API today, November 18 (UTC+8). As the operator, we state that all OpenAI groups on the site now support them; this check did not inspect every group one by one.

What OpenAI says they are

This is OpenAI’s positioning, not an independent review from us. Astra is the new frontier model for the hardest projects, with a focus on professional slides, spreadsheets and documents, plus cleaner code and stronger front-end work. Sol is the efficient, versatile model for everyday work; OpenAI says it is smarter and more efficient than GPT-6.1 Sol. Both official model cards list a 1,050,000-token context window, a 922,000-token max input, and a 128,000-token max output. Knowledge cutoffs are May 18, 2026 for Astra and June 1, 2026 for Sol.

The official migration guide recommends Astra for the most demanding tasks and Sol for complex work at lower cost. Do not set reasoning effort to none. For Chat Completions, omit temperature, top_p, logprobs and top_logprobs. Tool calling requires the Responses API.

Sources: OpenAI announcement · Astra model card · Sol model card · Using GPT-6.2 · Release index

OpenAI’s standard API list prices (per 1M tokens) are $8 / $0.40 cached / $40 output for Astra, and $2 / $0.10 / $10 for Sol, which keeps GPT-6.1 Sol pricing. These are official OpenAI rates, not Twinkle’s. This site’s prices are on the model plaza. Our page capture did not return unit prices, so this post quotes none.

A two-minute connectivity check

At about 09:06 UTC+8 on November 18, GET /v1/models returned HTTP 200 and included both IDs. We then sent the same short Chat Completions request to each model: reply with exactly ok, max_tokens=16, reasoning_effort="low", no temperature.

Both calls returned HTTP 200, the requested model ID, content ok, and finish_reason=stop. Curl total time was 21.195 seconds for Astra and 4.552 seconds for Sol. Both responses reported prompt_tokens=4393, completion_tokens=5, and total_tokens=4398. The prompt usage is far larger than the visible text and may include a system or reasoning preamble; we are not asserting a mechanism. This was a smoke test, not a quality, latency, or long-context benchmark.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://big-model.smart-agi.com/v1",
    api_key=os.environ["TWINKLE_API_KEY"],
)

r = client.chat.completions.create(
    model="gpt-6.2-sol",  # or "gpt-6.2-astra"
    messages=[{"role": "user", "content": "Reply with exactly: ok"}],
    reasoning_effort="low",
)
print(r.choices[0].message.content)
print(r.usage)
Enter fullscreen mode Exit fullscreen mode

Site: https://big-model.smart-agi.com

Pricing: https://big-model.smart-agi.com/model-plaza

Twinkle Model focuses on low-cost, stable access to GPT, Claude, and Chinese models, at roughly one-tenth the official list price where the plaza lists it. Check the plaza for current rates.

I operate the API relay Twinkle Model (https://big-model.smart-agi.com). Pricing is listed on the model plaza: https://big-model.smart-agi.com/model-plaza.

Top comments (0)