Artificial Intelligence
Anthropic launches Claude Haiku 5.5 and cuts prices
Anthropic released Claude Haiku 5.5, said it cut token prices and lifted benchmark scores against Haiku 4.5, and also reduced Sonnet 5.5 caching costs and added API credits.
- Anthropic released Claude Haiku 5.5 and described it as its fastest and most affordable small model.
- The company said Haiku 5.5 averaged about 75 percent less than Haiku 4.5, with prices up to 90 percent lower for prompts up to 100,000 tokens and five times higher for longer prompts.
- Anthropic said Haiku 5.5 scored 1,620 on GDPval-AA v2.1, 45.9 percent without tools and 57.4 percent with tools on Humanity's Last Exam, 72.4 percent on OSWorld-2.1, and 39.2 percent on Terminal-Bench 4.0.
- Anthropic said Haiku 5.5 was available on Amazon Web Services, Google Cloud, and Microsoft Azure, and that it had cut Sonnet 5.5 cache read costs from $0.20 to $0.10 per million tokens.
- Anthropic said the model used an updated tokeniser that consumed slightly more tokens per task, so the real savings would be smaller than the per-token pricing suggested.
Anthropic released Claude Haiku 5.5 and said it was its fastest and cheapest small model. The company said it was aimed at high-volume work such as summarisation, database queries, classification, and live customer support.
Anthropic said average pricing was about 75 percent lower than for Haiku 4.5. For prompts up to 100,000 tokens, which it said represented about 90 percent of earlier Haiku use, it said prices fell by as much as 90 percent, while longer prompts cost five times more.
The company said the model used an updated tokeniser that increased token use slightly, which would reduce the practical saving. It said a similar change in Opus 4.x raised token use by about 30 percent.
Anthropic said Haiku 5.5 scored 1,620 on GDPval-AA v2.1, 45.9 percent without tools and 57.4 percent with tools on Humanity's Last Exam, 72.4 percent on OSWorld-2.1, and 39.2 percent on Terminal-Bench 4.0. It also said Haiku 5.5 led OpenAI's GPT-6 Luna in the benchmarks it listed, while still trailing Sonnet 5.5 on some reference scores.
Anthropic said Haiku 5.5 was available on Amazon Web Services, Google Cloud, and Microsoft Azure. It also said it had halved Sonnet 5.5 cache read charges from $0.20 to $0.10 per million tokens and was adding monthly API credits for subscribers.
Named in this story
Companies
- Anthropic
- released Claude Haiku 5.5 and cut Sonnet 5.5 pricing
- OpenAI
- the rival whose budget model Anthropic compared against
- Amazon Web Services
- one of the platforms carrying the model
- Google Cloud
- one of the platforms carrying the model
Products and systems
- Claude Haiku 5.5
- the new small model Anthropic launched
- Haiku 4.5
- the earlier model used for comparison
- Sonnet 5.5
- the model whose cache read costs Anthropic reduced
- Opus 5.5
- the larger model Anthropic used as a higher-end comparison
- GPT-6 Luna
- OpenAI's budget model in the comparison
- Microsoft Azure
- one of the platforms carrying the model
- GDPval-AA v2.1
- one of the benchmarks Anthropic cited
- Humanity's Last Exam
- one of the benchmarks Anthropic cited
- OSWorld-2.1
- one of the benchmarks Anthropic cited
- Terminal-Bench 4.0
- one of the benchmarks Anthropic cited
How the source tells it
The register was promotional and competitive, with benchmark winner framing and a speculative explanation for the pricing move.
- boosterism superlatives and broad claims of benchmark superiority
- competitive escalation price-war language and repeated comparison with a rival model
- speculation as certainty an unqualified causal claim about the pricing cut