Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingAnthropic releases Claude Haiku 5.5: Cheaper, faster and more powerful mini model
On Wednesday, Anthropic released Claude Haiku 5.5, calling it the cheapest, fastest and most powerful mini model the company has built. The model is mainly targeted at highly concurrent tasks such as document summaries and database queries, but also suitable for speed-sensitive tasks such as real-time customer support and operating web browsers for users.
For developers who integrate Claude into their products, if the prompt word length is less than 100,000 tokens, the rate for input tokens is US$0.10 per million, and the rate for output tokens is US$0.50 per million. Token is the basic unit of text read and written by AI models, which is equivalent to approximately three-quarters of a word. It is also the basic unit of billing by large AI companies. This pricing is in line with the rate of the rival small model GPT-6 Luna, which OpenAI released on September 22.
Previously, Haiku 4.5 rates were US$1 and US$5, so the new rate was reduced by 90%; while for prompts exceeding 100,000 tokens, the fee was reduced by 50%. Because about 90% of requests for older models are within this limit, and Haiku 5.5 splits text into slightly more tokens, Anthropic estimates that the average savings are close to 75%.
Customer support chats and long email summaries were exactly the original intention of Anthropic developing Haiku. Therefore, tasks that are inherently repetitive, simple and non-creative are the best scenarios for this model.
That being said, benchmarks show that this is by no means a stupid model. For example, in the OSWorld 2.1 test (which evaluates whether AI can operate a real computer through long-term multi-step tasks, with scores calculated as partial credit percentages), Haiku 5.5 performed better than OpenAI's GPT Luna, with success rates of 72.4% versus 48.9%.
In Terminal-Bench 4.0, AI agents are required to complete professional tasks by typing their own commands, with scores based on the proportion of correct attempts on the first time. Haiku 5.5 scored 39.2%, while OpenAI's Luna scored 16.4%, and Haiku 4.5 scored 0%. For comparison, Anthropic's Claude Sonnet 5.5 score was 70.6%.
We tested the model on a simple logic problem, and the response speed was extremely fast, almost instantaneous. However, it gives the wrong answer, which is not what users want from the model, so don't blindly trust its answers.
In the GDPval-AA v2.1 test (which uses the Elo rating system derived from chess to rate models based on real professional work in 44 professions), Haiku 5.5 scored 1620 points. Luna scored 1437 points, while Haiku 4.5 only scored 735 points.
This is also the first Haiku model with an adjustable "effort level" setting. This setting allows users to balance cost with smarter answers. In addition, Anthropic has halved the cache read price of Sonnet 5.5 to $0.10 per million tokens, a discounted rate for text that has been processed by the model.
Haiku 5.5 的发布紧随 Opus 5.5（9 月 22 日发布）15 天之后，以及 Sonnet 5.5（9 月 28 日发布）9 天之后，后者是 Anthropic 承诺推出的 Claude 5.5 系列中的最后一款。Opus 5.5 是自 CEO Dario Amodei 发表文章呼吁行业放缓 AI 能力提升步伐以来的首次发布。
目前，Claude Haiku 5.5 已在 Claude 网站、Amazon Web Services、Google Cloud 和 Microsoft Azure 上线，模型名称为 claude-haiku-5-5。Anthropic 本周还在推出月度 API 额度优惠：Max 5x 计划订阅者可获得 100 美元，Max 20x 计划为 200 美元，Team 计划用户则可共享高达 500 美元的额度。
