88Z AI, 8 October — Anthropic has launched its latest lightweight model, Claude Haiku 5.5, which outperforms DeepSeek V4.1 Flash and GLM-5.3-Flash in benchmarks. Positioned as a competitor to GPT-6 Luna, it offers a lower price point, with a 90% reduction in costs for inputs and outputs within 100,000 tokens. The model is the first to support five levels of effort adjustment, achieving a 72.4% score in the Max setting of the OSWorld 2.1 computer operation test. While powerful, Haiku 5.5 does not replace Sonnet 5.5
At a glance
- Claude Haiku 5.5 outperforms rivals with a 90% cost drop for under 100k tokens
- First to offer five effort levels, targeting execution tasks rather than replacing Sonnet
The update
Anthropic has officially released its latest lightweight model, Claude Haiku 5.5. In terms of performance, the model's benchmark scores have surpassed those of DeepSeek V4.1 Flash and GLM-5.3-Flash. It aims to compete with GPT-6 Luna while offering a more attractive pricing strategy. Specifically, the price for inputs and outputs within 100,000 tokens has been reduced by 90%. Furthermore, Claude Haiku 5.5 is the first model to support five levels of effort adjustment. Users can achieve a score of 72.4% in the Max setting of the OSWorld 2.1 computer operation benchmark. Despite these enhanced capabilities, it is noted that the model still cannot replace Sonnet 5.5, as it is primarily focused on handling execution-level tasks and serving as a sub-agent.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 量子位Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.