Architect designing a complex city model and professional sketching notes, showing varied AI applications

GPT-6.2 Astra and Sol

OpenAI has launched GPT-6.2 Astra and GPT-6.2 Sol, two new closed models. Astra is designed for difficult, long-horizon tasks, while Sol is for fast, versatile daily use. Both models offer improved cost-effectiveness for developers.

OpenAI released GPT-6.2 Astra for complex professional work and GPT-6.2 Sol for efficient everyday tasks, both available in the API.

Climber on a difficult rock face and person on a scooter in a city, representing Astra's and Sol's uses

TL;DR

OpenAI has launched GPT-6.2 Astra and GPT-6.2 Sol, two new closed models. Astra is designed for difficult, long-horizon tasks, while Sol is for fast, versatile daily use. Both models offer improved cost-effectiveness for developers.

Key facts

  • Maker: OpenAI
  • What's new: New frontier model (Astra) and efficient everyday model (Sol)
  • Access: Available in the API as gpt-6.2-astra and gpt-6.2-sol
  • Price: Astra input $8/MTok, output $40/MTok; Sol input $2/MTok, output $10/MTok
  • Context: 1,050,000-token context window
  • Performance: Astra scores 62 on Artificial Analysis Intelligence Index
Hand placing a chess piece on a long path, next to coins, symbolizing model access, context, and cost

What is it?

GPT-6.2 Astra and GPT-6.2 Sol are OpenAI's latest closed models. Astra is a frontier reasoning model for complex professional work, succeeding GPT-6 Astra. Sol is an efficient, versatile model for everyday tasks, replacing GPT-6.1 Sol.

How does it work?

Both models accept text and image input and return text. They share the same context limits, with Astra having a knowledge cutoff of May 18, 2026, and Sol of June 1, 2026. They are designed to improve cache reuse and preserve reasoning across turns.

Why does it matter?

These models offer frontier results at a lower cost per task, with Astra's token prices falling 20%. Astra also meets the High threshold for AI self-improvement. However, coding gains did not improve, and hallucination rates remain high.

Who is it for?

Developers and professionals working on hard, long-horizon tasks or fast, versatile everyday tasks.

Try it

Available in the OpenAI API as gpt-6.2-astra and gpt-6.2-sol

Frequently asked questions

How much does GPT-6.2 cost in the API?

GPT-6.2 Astra's input price is $8 per million tokens and output is $40 per million tokens. GPT-6.2 Sol's input price is $2 per million tokens and output is $10 per million tokens. Cached input for Astra is $0.40 per million tokens.

What are the context limits for GPT-6.2 Astra and Sol?

Both GPT-6.2 Astra and Sol ship with a 1,050,000-token context window. They have a max input of 922,000 tokens and a max output of 128,000 tokens.

How does GPT-6.2 Astra perform compared to other models?

GPT-6.2 Astra scores 62 on the Artificial Analysis Intelligence Index, placing it second behind Claude Fable 5.2 (65). It ranks #1 on AA-3D with 1,318 Elo, ahead of Fable 5.2 (1,271).

What are the hallucination rates for GPT-6.2 Astra and Sol?

GPT-6.2 Astra posts a 59% AA-Omniscience hallucination rate. GPT-6.2 Sol has a 55% AA-Omniscience hallucination rate. Hallucination rates remain high for both models.

Sources

  1. Simonwillison
  2. MarkTechPost