Hajra Naz
Anthropic’s newest model, Opus 5.5, was released Tuesday. According to the company, it sets a new state-of-the-art benchmark. This applies to both coding and knowledge work performance. Opus represents the most capable and expensive tier in Anthropic’s lineup. That lineup has three tiers total. Sonnet sits in the middle.
Haiku remains the fastest and cheapest option. Notably, Anthropic says this release outpaces its larger Fable model in many benchmarks. It also succeeded on several informal tasks that Fable had previously failed to complete.
This new model is also significantly cheaper than its predecessor. Output tokens will cost $20 per million tokens for Opus 5.5. That compares to $25 for the previous model. Other pricing metrics show similar drops too. The model also runs faster overall. This reflects a broader drop in the compute required to serve it.
The new version brings significant changes to how Opus communicates. Opus 5.5 uses less jargon than before. It’s also more likely to place important information at the start of its messages now.
Read More: Claude Opus 5 Just Launched: Here’s Why Everyone Is Talking About It
This launch comes just two months after Opus 5 was first released. That earlier version launched on July 24. According to the announcement, both Sonnet 5.5 and Haiku 5.5 will follow soon. These represent the next tiers in the lineup. Anthropic says they’ll launch “in the coming weeks.” Both will feature similar performance improvements.
Anthropic notes that Opus 5.5 is comparable to Mythos in specific capabilities. That includes biology and cybersecurity knowledge. Because of this, its release follows the same safeguards as Anthropic’s Fable model. These safeguards limit certain model uses. That includes discovering exploits in compiled programs. It also includes developing recognizable biological weapons, among other restricted tasks.
Opus 5.5 marks Anthropic’s first model release since a notable shift. CEO Dario Amodei recently embraced calls to “pace the frontier.” This means deliberately slowing AI capability progress. The goal is matching that pace to progress on alignment research.
“I have become convinced that fully addressing the risks requires even more prudence,” Amodei wrote in a post earlier this month. “Not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”
Read More: OpenAI Introduces GPT-6 Sol and Luna With Fewer Mistakes and Lower Costs
Opus 5.5’s safety training remained broadly similar to previous models. This included alignment testing. It also included pre-release evaluation by outside organizations. That includes groups like METR and Frontier Design. Still, Anthropic emphasized something important. More advanced training and evaluation systems are already being prepared for future models. That includes improved security and monitoring systems too.
“As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe. That capacity takes time to build, and we’ve started to put the infrastructure in place to support it,” the blog post reads. “We expect to share more details on these efforts soon.”
