Skip to main content
The Quantum Dispatch
Back to Home
Cover illustration for GPT-6.2 Astra and Sol: Prices, Scores and Who Should Switch

GPT-6.2 Astra and Sol: Prices, Scores and Who Should Switch

GPT-6.2 Astra drops to $8 per million input tokens as Sol holds at $2. Here is what changed, what independent tests show and which model fits your work.

Dr. Nova Chen
Dr. Nova Chen★Nov 18, 2026★4 min read

OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, and the most interesting number in the announcement is not a benchmark. It is the cost of finishing a task. GPT-6.2 Astra is the new frontier model for demanding professional work, GPT-6.2 Sol is the efficient everyday option, and both are designed to reach a good result with fewer output tokens. For teams choosing a model this week, the question is less about which name sits on top of a chart and more about where the extra capability justifies the price.

  • Astra API price: $8 per million input tokens, $0.40 cached input and $40 output, with input and output 20% below GPT-6 Astra.
  • Sol API price: $2 input, $0.10 cached input and $10 output per million tokens, unchanged from GPT-6.1 Sol.
  • Independent score: Artificial Analysis rates Astra 62 and Sol 57 on its Intelligence Index, with Claude Fable 5.2 at 65, according to Decrypt.
  • Availability: Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, plus the OpenAI API, from November 17, 2026.

What Changed in GPT-6.2 Astra?

OpenAI's emphasis is on professional deliverables. The company says GPT-6.2 Astra follows slide templates and spreadsheet conventions more closely, writes more directly, produces cleaner code and handles computer use with enough precision to work on individual parts in CAD and simulation tools.

The vendor's benchmark claims are ambitious, so they deserve clear attribution. OpenAI reports that Astra scores 5.2 points above GPT-6 Astra on Agents' Last Exam at less than a third of the cost, and 6.2 percentage points higher on its internal GDPval v2 test while using 22% fewer output tokens. It also says Astra leads Claude Fable 5.2 on several task-level comparisons at a much lower cost per task. These are OpenAI's figures, and one footnote matters: the Agents' Last Exam leaderboard result for Fable 5.2 excludes nine tasks that used a fallback model.

How Do Independent Benchmarks Compare?

Third-party results paint a closer race. Decrypt reports that Artificial Analysis places GPT-6.2 Astra at 62 on its Intelligence Index, five points above the original GPT-6 Astra and just behind Claude Fable 5.2 at 65. Sol scores 57, matching the original Astra at roughly 15% of its cost per evaluated task. Decrypt also notes no gain for Astra in coding-agent performance in that testing.

That makes the practical story one of efficiency. Our earlier GPT-6.1 Sol and Astra pricing comparison showed how much routing matters, and this release strengthens the case for sending most work to the cheaper tier.

Should You Use GPT-6.2 Astra or Sol?

Start with Sol for agent loops, drafting, support tasks and ordinary coding. OpenAI says Sol came within 0.5 points of Astra on Cognition's FrontierCode 2.0 at about one-fifth of the cost. Reserve Astra for long, high-stakes work such as financial models, legal analysis, scientific tasks and detailed design. For API users, OpenAI recommends keeping prompt prefixes unchanged and appending new instructions, which helps the discounted cache rate apply.

GPT-6.2 Safety Ratings and Safeguards

OpenAI classifies both models as Critical for cybersecurity and High for biological and chemical capability under its Preparedness Framework. GPT-6.2 Astra is also its first externally released model rated High for AI self-improvement, and the company says it added stronger containment and a monitoring system called Control Room. The tighter cyber safeguards can restrict some legitimate security research, and GPT-6.2 has not yet reached OpenAI's Daybreak program for vetted defenders.

Decrypt also observed that the models had not yet appeared on its Plus account shortly after launch, so a gradual rollout is possible. For more model analysis, visit our AI coverage.

Sources: OpenAI — November 17, 2026; Decrypt — November 17, 2026; 9to5Google — November 17, 2026.

More AI Stories