Anthropic, OpenAI launch dueling models in AI price war

The generative AI industry's two leading labs released competing models within hours of each other on Tuesday, each built around the same pitch: frontier-level intelligence should cost less.

Anthropic launched Claude Opus 5.5, its strongest mid-tier model to date, while OpenAI introduced GPT-6 Sol and GPT-6 Luna with API prices cut 50 percent from their predecessors. The near-simultaneous releases mark a shift in the AI competition from a pure capability race toward an economics war, as both companies vie for cost-conscious enterprise customers and face growing pressure from cheaper Chinese alternatives.

Dueling Price Cuts

Anthropic's Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 respectively for Opus 5. The company says the model delivers performance comparable to its top-tier Fable 5.1 while costing 40 percent less to run than its predecessor, thanks to both lower per-token prices and reduced token usage per task.anthropic+2

Roughly 90 minutes later, OpenAI announced GPT-6 Sol at $2 per million input tokens and $10 per million output tokens, and GPT-6 Luna at just $0.10 and $0.50 respectively. OpenAI attributed the 50 percent reductions to improvements in inference efficiency and prompt caching, which it said can cut costs on cached input reads by up to 90 percent.openai

"We're passing those savings directly on to users and customers," OpenAI wrote in its announcement, framing Sol and Luna as extensions of the technology behind GPT-6 Astra, its flagship model released earlier this month.openai

Why Cost Per Task Now Matters More

The pricing moves reflect a broader industry recalibration. As AI shifts from answering occasional prompts to powering sustained agentic workflows across thousands of interactions, token economics have moved from abstract to material. OpenAI highlighted this explicitly: on its AutomationBench test, GPT-6 Sol outperformed Claude Opus 5 at roughly 9 percent of Opus 5's cost per task.openai

Fortune ↘ 2.96% reported that both companies are battling not only each other but also lower-cost open-weight models and Chinese rivals including Moonshot and DeepSeek, which have drawn cost-conscious users away from American labs. The Financial Times reported earlier this year that prices customers pay for models from leading U.S. labs had already fallen nearly a quarter since mid-July.ft+1

A New Competitive Landscape

The dueling launches suggest the AI industry's next phase will be defined less by which lab builds the smartest model and more by which can deliver sufficient intelligence at the lowest cost. Anthropic continues to maintain specialized systems above Opus for demanding work, and OpenAI still positions Astra as its most capable offering. But for most enterprise workloads, the question has changed.anthropic+1

"The argument is no longer simply: ours is smarter," the tech publication We Are Social Media wrote Tuesday. "It is: ours gets enough of the job done for less".wersm