Skip to main content
OpenAI's GPT-6.2 Astra, represented by a glowing abstract neural network, leads new AI models with a 1,318 Elo score.

Editorial illustration for OpenAI's GPT-6.2 Astra leads new AI models with 1,318 Elo score

OpenAI's GPT-6.2 Astra Tops AI Rankings at 1,318 Elo

• 4 min read

OpenAI put out two new models Tuesday, GPT-6.2 Astra and GPT-6.2 Sol, and priced them to undercut the competition rather than just beat it on raw scores. Astra is built for the heavy lifting: legal research, financial modeling, scientific work that needs precision over speed. Sol handles the rest, the quick back-and-forth tasks most people actually run through ChatGPT all day. Both are live now for Plus, Pro, Enterprise, and Edu users inside ChatGPT Work and Codex, and developers can pull them through the API starting immediately.

The pricing move is the part worth sitting with. Astra's API costs dropped 20 percent from its predecessor, landing at $8 per million input tokens and $40 per million output tokens. Sol didn't move at all, holding steady at GPT-6.1 Sol's old rates.

OpenAI also expanded access to its Ultrafast mode, giving Pro-200 subscribers a tier previously reserved for Pro-500 users, starting with Sol. Benchmark results from Artificial Analysis put Astra in a tight race with Anthropic's top model, and that comparison is where things get interesting.

OpenAI owns the Pareto frontier on cost versus performance, the line where no other model is both cheaper and better, according to Artificial Analysis.

Why this matters

For developers and founders, the real headline isn't the 1,318 Elo score, it's the 20 percent price cut on Astra's API alongside a 9-point jump on Terminal-Bench-Science at less than a third of the old cost. That combination changes the build-versus-buy math for anyone running scientific or financial workflows through an API. Claude Fable 5.2 still sits close behind at 1,271, so Anthropic hasn't lost the plot, but OpenAI pairing a capability gain with a price drop narrows the room competitors have to justify premium pricing.

We'd watch what Anthropic does next on cost rather than benchmarks alone. Researchers should treat the Terminal-Bench-Science number as a useful signal but not a verdict: one benchmark, one model generation, no independent replication mentioned here. The Astra/Sol split, heavy-duty versus everyday, also tells us OpenAI is betting customers will pay for routing rather than one giant model.

Worth tracking whether that bet shows up in actual enterprise adoption numbers over the next few quarters, not just leaderboard chatter.

Common Questions Answered

What are the key differences between GPT-6.2 Astra and GPT-6.2 Sol?

GPT-6.2 Astra is designed for complex, precision-focused tasks such as legal research, financial modeling, and scientific work that prioritize accuracy over speed. GPT-6.2 Sol handles quick, conversational back-and-forth tasks that most users perform daily through ChatGPT, making it optimized for general-purpose interactions.

What does it mean that OpenAI owns the Pareto frontier on cost versus performance?

According to Artificial Analysis, OpenAI's position on the Pareto frontier means no other AI model is both cheaper and better than their offerings simultaneously. This indicates OpenAI has achieved an optimal balance where competitors cannot match their combination of lower pricing and superior performance metrics.

How much did OpenAI reduce API prices for GPT-6.2 Astra?

OpenAI implemented a 20 percent price cut on Astra's API while simultaneously achieving a 9-point jump on Terminal-Bench-Science benchmarks. This price reduction at less than a third of the old cost significantly changes the economic calculation for developers deciding whether to build custom solutions or use OpenAI's API for scientific and financial workflows.

Which platforms and user types have access to GPT-6.2 Astra and Sol?

Both GPT-6.2 Astra and Sol are available for Plus, Pro, Enterprise, and Edu users within ChatGPT Work and Codex. Additionally, developers can access these models through OpenAI's API for integration into their applications.

How does Claude Fable 5.2 compare to GPT-6.2 Astra in performance?

GPT-6.2 Astra leads with a 1,318 Elo score compared to Claude Fable 5.2's 1,271 Elo score, representing a 47-point performance advantage. However, the article notes that Anthropic's Claude remains competitive and close behind, indicating the AI model competition remains tight despite OpenAI's current lead.

LIVE05:12California Takes OpenAI to Court in New Lawsuit Claiming Private Ownership