Anthropic and OpenAI, Once Calling for Slowdown, Shift to Cost-Effective Speed Race With New AI Models

Anthropic and OpenAI, which previously raised concerns over AI risks and called for slowing down development, have unveiled new AI models centered on cost efficiency side by side.

Anthropic introduced its new AI model, Claude Opus 5.5, on the 22nd (local time).

This model delivers performance comparable to its highest-tier model, Claude Fable 5.1, while further lowering costs.

Introducing the model, Anthropic stated, "This is the first model we are releasing since we urged for a slowdown in cutting-edge AI," adding, "It performs at the level of Fable 5.1 in most tasks while costing 40% less than its predecessor, Opus 5."

In particular, it received higher evaluations than the top-tier model in several areas, recording higher scores than Fable 5.1 in three coding-related benchmarks—TerminalBench, FrontierCode, and CursorBench—as well as knowledge-work benchmarks.

Anthropic explained that when tested across 2,000 scenarios in "alignment," which ensures AI sets goals aligned with human intentions, Opus 5.5 recorded the highest score among all Claude models released to date.

The company also noted that safety measures regarding sensitive fields such as biology and cybersecurity, as well as "distillation" used for copying AI models, were raised to a level similar to Fable 5.1 to reduce risk factors.

On the same day, OpenAI also joined the cost-efficiency competition by unveiling GPT-6 Sol and GPT-6 Luna, lower-tier models of its flagship GPT-6 Astra.

OpenAI stated, "If GPT-6 Astra showcased a new generation of intelligence, Sol and Luna expand the horizons of cost efficiency to help spread the benefits of intelligence."

OpenAI engaged in a more aggressive pricing competition, offering API (Application Programming Interface) fees for GPT-6 Sol and Luna at half the price of the previous GPT-5.6 Sol and Luna.

Artificial Analysis, an AI performance evaluation firm, assigned Claude Opus 5.5 an intelligence index (AAII) of 58 points, higher than Fable 5.1, placing it in the number one spot.

GPT-6 Sol scored 48 points, tying for fifth place.

The two companies' competing releases of new models emphasizing cost efficiency are interpreted as a strategy to defend their market dominance while previously advocating for a "slowdown."

By elevating the performance of secondary-tier models to a level comparable to top-tier ones—such as Claude Mythos and Fable, and GPT Astra—and lowering prices, the strategy aims to prevent customers from defecting to latecomers or open-source models from China.

It is also highly likely that calculations were made to provide high-value models to the market to boost revenue, as both companies prepare for initial public offerings (IPOs).
※ Please note: This article was translated by AI and may contain errors.