OpenAI and Anthropic Ship Cheaper, Faster Models Same Day

OpenAI and Anthropic both released new models on Tuesday, and both pitched the same selling point: more speed and lower cost than their predecessors.

Anthropic's new model, Claude Opus 5.5, matches the capabilities of Fable 5.1 while cutting cost by nearly 40% compared to Opus 5. OpenAI's GPT-6 Sol and GPT-6 Luna follow the 5.6 versions released in July, and arrive just weeks after ChatGPT Astra, which launched in early September.

The simultaneous releases show where the frontier-lab competition has shifted. Raw capability gains are no longer the only battleground; pricing, usage limits and efficiency now dominate the pitch to business and developer customers who run models at scale.

GPT-6 Sol and Luna

OpenAI designed the two new models to make the power of GPT-6 Astra accessible at lower budgets, with higher usage limits for business and developer workflows. Sol and Luna are trained with methods similar to Astra's, so users keep most of the benefits of the more powerful model, including improvements in coding and computer use. OpenAI still calls GPT-6 Astra the "world's best model" for computer use.

The new models also emphasize factuality — how useful the answers are after a prompt. OpenAI says its latest models make less than half the mistakes of their predecessors.

Efficiency in speed and cost is the core proposition. OpenAI shared benchmarks for the pair, but the gains likely won't translate much for everyday users.

Both GPT-6 Sol and Luna are rolling out to ChatGPT Work and Codex. Free users and lower-tiered subscribers will still get access to GPT-6 Luna.

(Disclosure: Ziff Davis, CNET's parent company, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Claude Opus 5.5

Anthropic is touting Opus 5.5 as its strongest-performing model to date on alignment — the degree to which a model matches human intentions, values and rules. The claim carries weight after AI agents made headlines this summer by completing tasks by any means necessary. Anthropic's statement doesn't mean the model is perfect; its agentic behavior will need real-world testing.

Opus 5.5 ships with the same safeguards Anthropic developed for its most capable models. That means it may revert to an older, less advanced model for certain cybersecurity or scientific tasks.

The model also requires less compute than Opus 5. In Anthropic's own testing at default settings, the new model costs 40% less on typical workloads. Anthropic is additionally increasing five-hour usage limits on Pro, Max and Team plans.

Anthropic said the new models communicate more naturally than their predecessors. Writing and instruction-following improved in tests — a change the company took directly from user feedback.

The back-to-back releases set up a pricing race for the rest of the year, with both labs now betting that cheaper inference, not just benchmark wins, will win enterprise customers.