OpenAI and Anthropic Launch "Low-Cost Models" on the Same Day, Kicking Off a "Price War"

On the same day, two top AI labs in the United States successively launched lower-priced versions of their flagship models, signaling that the AI industry’s competition has shifted from a pure arms race of capabilities to a dual contest of price and performance.

On Tuesday, Anthropic released its new generation model Claude Opus 5.5, with usage costs reduced by approximately 40% compared to the previous version; OpenAI on the same day launched two lower-priced versions of ChatGPT-6, Sol and Luna, with prices 50% lower than the ChatGPT-5.6 series. The near-simultaneity of these moves by both companies, with the timing almost overlapping, clearly outlines the fierce competition currently playing out in the AI market.

Behind this wave of price cuts lies continued pressure from Chinese "open-weight" models—which allow users to customize and run them on local hardware. Meanwhile, Anthropic is actively preparing for its planned IPO this fall, which is widely expected to be the largest in history; OpenAI, after a sluggish start to the year, is hoping that price reductions will reignite its growth momentum.

Pricing Details: The Pricing Landscape of Anthropic and OpenAI

In terms of specific pricing, Anthropic’s Opus 5.5 charges $4 per million input tokens and $20 per million output tokens, while its current strongest model Fable 5.1 charges $10 and $50 for the same, respectively—on par with the pricing of OpenAI’s flagship model Astra. According to Anthropic, Opus 5.5 matches Fable 5.1’s performance on most tasks.

For OpenAI, Sol is priced at $2 per million input tokens and $10 per million output tokens; the lightweight version, Luna, goes as low as $0.10 per million input tokens and $0.50 per million output tokens. Earlier this month, OpenAI took the lead in launching its strongest model, Astra, with Sol and Luna following as budget options within that lineup.

Reports indicate that after a period of all-out arms race focused on maximizing capability, the two major US AI labs are increasingly shifting their competitive focus to a comprehensive contest of price and performance, aiming to win over enterprise clients that are sensitive to costs.

Chinese "open-weight" models are becoming an increasingly notable competitive threat. According to cloud application company Vercel, these customizable and locally deployable models are steadily capturing market share from US labs. Under this pressure, OpenAI has cut model prices multiple times in recent months and claims its Astra model now offers the best performance on the market.

OpenAI previously stated that it is unlikely to go public before 2027. After growth slowed at the start of the year, lowering prices became one of its core strategies to reactivate expansion.

For Anthropic, the significance of this release goes beyond price competition. The company is aggressively expanding its client base to pave the way for its planned IPO this fall—which is anticipated to become one of the largest tech IPOs in history. Investors widely speculate that the launch of the new model will act as a catalyst for the company to formally submit its IPO prospectus.

A central question closely watched by the market: as competition intensifies and clients become increasingly sensitive to pricing, can Anthropic sustain its high growth trajectory over the coming years?

Safety Commitment: Emphasizing Risk Management Alongside Price Cuts

Notably, this release also comes at a time when AI safety issues are drawing growing attention.

Opus 5.5 is the first new model launched by Anthropic after CEO Dario Amodei publicly called on AI labs to "keep pace with frontier development" and to deeply integrate third-party testing organizations. Previously, researchers have warned that an unconstrained race to develop self-improving superintelligent AI could pose risks to humanity.

Anthropic noted that the new model has undergone "extensive alignment testing, external evaluations before release, and the implementation of corresponding safeguards in high-risk fields such as cybersecurity and biology."

OpenAI on Tuesday also announced new priorities for further cooperation with independent evaluation organizations, covering areas such as model training, safety protocols, risk preparedness, and incident investigation. Just a week prior, Anthropic had revealed a partnership with Accenture to conduct safety tests.