Anthropic has released Claude Opus 5, its latest AI model. The company says it outperforms the competing Fable 5 model on key benchmarks while costing half as much. It also uses one-seventh the reasoning tokens of its predecessor, a leap in efficiency that could reshape how developers choose between large language models.
Half the cost, better performance
Claude Opus 5’s price tag is set at 50% below Fable 5’s, according to Anthropic. The company claims the new model achieves higher scores on standard AI benchmarks than Fable 5, though it did not release specific numbers. The combination of lower cost and better performance is rare in the fast-moving AI market, where more capable models often come with higher price tags.
For developers and enterprises running large-scale inference workloads, the pricing change is immediate. A company processing millions of queries a month could see its AI bill cut in half simply by switching. Anthropic did not say whether it plans to offer volume discounts or special pricing for academic researchers.
Reasoning efficiency boost
The reduction in reasoning tokens — to one-seventh of the previous Claude Opus model — means the model can produce the same quality of output with far less computational overhead. That matters for latency-sensitive applications like real-time chatbots, code assistants, and automated analysis tools. Fewer tokens also mean less energy consumed per task, a factor that’s drawn growing attention from corporate sustainability officers.
Anthropic didn’t name the predecessor model, but the improvement is substantial. The company has been working on optimizing its chain-of-thought reasoning process, and this release appears to be the first major payoff. Users running Claude Opus 5 on standard hardware should notice faster response times even on complex prompts.
Market competition heats up
Claude Opus 5 lands at a moment when AI model makers are racing to balance capability and cost. Fable 5, developed by a rival firm, has been a popular choice for high-stakes reasoning tasks. Anthropic’s move directly challenges that position by offering a cheaper, more powerful alternative.
The launch also puts pressure on other model developers — those with names like GPT, Gemini, and Llama — to justify their pricing. If Claude Opus 5 holds up in independent benchmarks, expect competitors to either cut prices or release new versions quickly. Some may try to match the reasoning token efficiency through architectural changes or specialized hardware.
Anthropic hasn’t disclosed how long it trained Claude Opus 5 or what hardware it used. The company has previously emphasized safety research and constitutional AI, but this release is squarely about raw performance and cost.
Claude Opus 5 is available now through Anthropic’s API. The company has not announced a date for open-source release or a local model variant.
