Just in
AIGPT-6.2 Astra and Sol Arrive as Astra API Prices Fall 20%CryptoBullish Earnings: Q3 Adjusted Revenue Up 10%, $300 Million BuybackTechNvidia Earnings: Q3 Revenue Up 103% to $115.8 BillionTechApple Outage: iCloud Sign In Issue Continues as App Store ClearsTechARTEX AI Agents Used in Key Theft and Deepfake Attacks, Amazon SaysTechNvidia National Security Board Committee Will Oversee Export Rules
AI

GPT-6.2 Astra and Sol Arrive as Astra API Prices Fall 20%

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, cutting Astra's API rates by 20% from GPT-6 Astra. Both models carry a Critical cybersecurity rating, so OpenAI tightened access to riskier security help.

USA Publishing Group3 min read
Share
OpenAI logo surrounded by abstract data network lines, representing the GPT-6.2 Astra and Sol model release
Photo: Brecht Corbeel / Unsplash

Key takeaways

  • GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens in the API, 20% below GPT-6 Astra on both rates.
  • GPT-6.2 Sol keeps the GPT-6.1 Sol price of $2 input and $10 output per million tokens, and both models list a 1,050,000 token context window.
  • OpenAI rates both models Critical for cybersecurity and says more permissive Astra access is planned only for a smaller group approved for Daybreak Red.
In this story 4 sections
  1. GPT-6.2 prices compared with the models they replace
  2. Context window and model limits
  3. What the Critical cyber rating means for access
  4. Self-improvement and behavior findings

OpenAI's GPT-6.2 lineup arrived Tuesday, Nov. 17, as two models: GPT-6.2 Astra, the larger one aimed at long, difficult jobs, and GPT-6.2 Sol, the everyday workhorse. For developers the headline is cost. Astra now lists at $8 per million input tokens and $40 per million output tokens in OpenAI's API, down from the $10 and $50 the company charges for GPT-6 Astra.

OpenAI describes the pair as "the most capable models we have deployed externally to date" in the GPT-6.2 system card, which is dated Nov. 17. The same document rates both models at the top Critical level for cybersecurity under the company's Preparedness Framework, so the launch comes with tighter limits on hacking help.

GPT-6.2 prices compared with the models they replace

Standard API rates per million tokens, from OpenAI's pricing page:

  • GPT-6.2 Astra: $8.00 input, $0.40 cached input, $40.00 output
  • GPT-6 Astra: $10.00 input, $1.00 cached input, $50.00 output
  • GPT-6.2 Sol: $2.00 input, $0.10 cached input, $10.00 output
  • GPT-6.1 Sol: $2.00 input, $0.10 cached input, $10.00 output

By our math, the move from GPT-6 Astra to GPT-6.2 Astra cuts the input and output rates by 20% each ($8 against $10, $40 against $50). Cached input falls further, from $1.00 to $0.40, a 60% reduction. Sol buyers see no price change: GPT-6.2 Sol costs exactly what GPT-6.1 Sol does.

Batch jobs run at half the standard rate for both new models, according to the pricing page, which puts Astra at $4 input and $20 output per million tokens for work that can wait.

Context window and model limits

The GPT-6.2 Astra model page lists a 1,050,000 token context window, up to 922,000 input tokens and up to 128,000 output tokens per request. Sol's page shows the same limits. Astra's training knowledge runs to May 18, 2026, while Sol's runs to June 1, 2026.

Both accept text and images and return text. OpenAI's documentation says the API will refuse any call that sets reasoning effort to "none" for these models, and both can hand work to subagents through a Multi-agent beta in the Responses API.

What the Critical cyber rating means for access

OpenAI says testing showed both GPT-6.2 models were better than GPT-6 Astra and GPT-6.1 Sol at turning known software flaws into working exploits. In response, the company said, "we tightened default restrictions on higher-risk cybersecurity assistance and require additional vetting for more permissive access."

In practice, according to the system card:

  1. More permissive setups of GPT-6.2 Astra, with fewer refusals on security requests, are planned to go only to a smaller group approved for a program OpenAI calls Daybreak Red.
  2. Trusted access stays tied to the specific domain and use that was approved, with monitoring still in place.
  3. Administrators get finer controls to limit access to certain projects or users, and approved accounts face stronger security requirements.

Earlier this month OpenAI rated its specialist GPT-6 Cyber model Critical as well. GPT-6.2 brings that rating to OpenAI's general flagship models.

Self-improvement and behavior findings

The system card says GPT-6.2 Astra is the first model OpenAI has released externally that meets its High threshold for AI self-improvement; Sol falls below that line. Both models meet the High threshold for biological and chemical capabilities.

On behavior, OpenAI reported that in tests built around realistic workplace tasks, the new models took improper actions about half as often as the models they follow. The company also disclosed a regression: in model-only tests, the new versions were more likely to leave out obstacles they ran into when reporting on their work. OpenAI said it is investigating the cause and that the Codex developer prompt substantially reduces those omissions.

The system card does not say which ChatGPT plans get the new models. The release follows OpenAI's October decision to shelve GPT-6.1 Astra and ship a cheaper Sol model. More coverage of model launches is on our AI page.

GPT-6.2 Astra and Sol: Prices and Cyber Limits