GPT-6.2 Astra und GPT-6.2 Sol von OpenAI

GPT-6.2 Astra and Sol put to the test

Since 18 November 2026, ChatX runs GPT-6.2 Astra and GPT-6.2 Sol. They replace GPT-6 Astra and GPT-6.1 Sol, which OpenAI superseded with the new versions the day before. We tested both models directly in the chat with three everyday tasks and show you when the more expensive Astra is worth it.

What OpenAI changed

OpenAI describes GPT-6.2 Astra as its new frontier model for demanding projects and GPT-6.2 Sol as a versatile model for everyday work. Both versions share the same goal: better results with fewer output tokens.

  • Astra gets cheaper, on ChatX too: Input and output cost 20 % less than with GPT-6 Astra, cached input even 60 % less. ChatX passes this on directly: the same Astra answer now costs you 20 % fewer tokens. Sol keeps the price of GPT-6.1 Sol.
  • More capability: In the internal benchmark GDPval v2, with long tasks from consulting, finance and design, Astra scores 6.2 percentage points above GPT-6 Astra according to OpenAI while using 22 % fewer output tokens. Sol gains 5.3 points over GPT-6.1 Sol.
  • Sol close to Astra: On the coding benchmark FrontierCode 2.0, Sol is only 0.5 points behind Astra at about a fifth of the cost.
  • Better documents: Astra is meant to produce spreadsheets, slides and texts that follow templates and the conventions of a field more closely and sound less like AI.
  • Large context: Both models can process up to roughly one million tokens at once.

The figures come from OpenAI’s announcement of 17 November 2026. Independent tests across many tasks are still pending.

Our test in the ChatX chat

We gave both models the same three tasks, each with the default thinking level “Low” and a single run. The token figures are the ChatX tokens that were actually charged.

Task GPT-6.2 Sol GPT-6.2 Astra
Turn a note into a customer email 2.7 s · 350 tokens 3.6 s · 1,382 tokens
Word problem on a trader’s profit 3.2 s · 1,103 tokens, correct 3.6 s · 4,242 tokens, correct
Pomodoro timer as an HTML file 16.5 s · 3,112 tokens, works 16.4 s · 12,164 tokens, works

The two emails are almost equally good: a clear subject line, the new date, the voucher and the contact person, all in a few sentences. Astra writes a little more formally. Both models calculate the profit of 69.12 euros correctly, Astra with the more compact working. Both timers also work on the first try with start, pause and reset.

The clearest difference is the price. For the same results, Astra used about four times as many tokens as Sol in our test.

Sol or Astra?

When is Sol enough? For emails, summaries, research with web search, calculations and most coding tasks. Sol is fast and delivers very solid results at a quarter of Astra’s price.

When is Astra worth it? For long, nested tasks: analysing extensive documents, building spreadsheets with many dependencies, larger code projects or texts that must follow a template strictly. According to OpenAI, that is where Astra plays to its strengths.

Which thinking level? On ChatX, both models offer the levels Low, Medium and High. Start with Low and only go higher for tricky tasks, because each level costs more tokens.

And without an account? Sol and Astra require a free ChatX account. For quick questions without signing up, there is GPT-6 Luna.

What changes for you on ChatX

You don’t need to change anything: if you previously chose GPT-6.1 Sol or GPT-6 Astra, you automatically continue with the new versions, including in existing chats. As always, ChatX charges for the tokens the model actually uses. Because OpenAI made Astra cheaper, an Astra answer now costs you 20 % fewer tokens than before. As Astra also needs fewer output tokens according to OpenAI, you save a little more on many tasks. The price of Sol stays the same.

You can try both models directly in the ChatGPT chat on ChatX: pick the model at the top of the input field and get started.


Posted

in

by

Tags: