Anthropic has launched Opus 5.5, its latest model that matches the performance of Fable in many benchmarks while being 20% cheaper than its predecessor. The release also introduces improved communication and safety features.
TL;DR
- Anthropic's Opus 5.5 outperforms Fable in many benchmarks and is 20% cheaper than Opus 5.
- The model is faster to run and uses less compute, with enhanced communication and safety features.
- Sonnet 5.5 and Haiku 5.5 are expected to be released in the coming weeks with similar improvements.
What happened
Anthropic released Opus 5.5 on Tuesday, setting a new state-of-the-art in coding and knowledge work performance. The model outperforms the larger Fable model in many benchmarks and succeeds in tasks that Fable failed to complete. Opus 5.5 is also significantly cheaper than its predecessor, with output tokens charged at $20 per million tokens, compared to $25 for Opus 5. Other metrics have similar price drops. The model is also faster to run, reflecting an overall drop in the compute required to serve it.
Opus 5.5 introduces significant changes to how the model communicates, with a reduced likelihood of using jargon and a greater emphasis on putting important information at the start of its messages. The launch comes just two months after the release of Opus 5 on July 24. According to the announcement, Sonnet 5.5 and Haiku 5.5, which are the next tiers in the lineup, will be released "in the coming weeks," with similar performance improvements.
Why it matters
The release of Opus 5.5 is significant for developers and startups looking for high-performance AI models at a lower cost. The model's enhanced communication features and reduced compute requirements make it a more efficient and user-friendly option. The upcoming releases of Sonnet 5.5 and Haiku 5.5 with similar improvements will provide more options for users at different price points.
Anthropic's decision to pace the frontier and deliberately slow down progress on AI capabilities to match the rate of progress on alignment is a notable development. The company's emphasis on safety and the introduction of more advanced training and evaluation systems for future models highlight the importance of risk prevention in AI development. The upcoming details on these efforts will be crucial for investors and policymakers.
Key facts
- Opus 5.5 outperforms Fable in many benchmarks and succeeds in tasks that Fable failed to complete.
- Output tokens for Opus 5.5 are charged at $20 per million tokens, compared to $25 for Opus 5.
- The model is faster to run, reflecting an overall drop in the compute required to serve it.
- Opus 5.5 introduces significant changes to how the model communicates, with a reduced likelihood of using jargon and a greater emphasis on putting important information at the start of its messages.
- Sonnet 5.5 and Haiku 5.5 are expected to be released in the coming weeks with similar performance improvements.
- Opus 5.5 is subject to the same safeguards as the company's Fable model, limiting its use in certain tasks.
- The model's safety training was broadly similar to its predecessors, with alignment testing and pre-release evaluation by outside organizations like METR and Frontier Design.
- Anthropic emphasized that more advanced training and evaluation systems are being prepared for future models, including improved security and monitoring systems.
Context
Anthropic's release of Opus 5.5 comes at a time when the AI industry is increasingly focused on the balance between capabilities and safety. The company's decision to pace the frontier and invest in risk prevention aligns with broader industry trends towards responsible AI development. The upcoming releases of Sonnet 5.5 and Haiku 5.5 with similar improvements will provide more options for users at different price points, further expanding the market for AI models.
The emphasis on safety and the introduction of more advanced training and evaluation systems for future models highlight the importance of risk prevention in AI development. Anthropic's efforts in this area are likely to influence other companies in the industry, leading to a more collective approach to addressing the risks associated with AI.
