Anthropic releases Haiku 5.5 with 90% cost cut, beating GPT-6 Luna on benchmarks
Anthropic released Haiku 5.5, its cheapest and fastest AI model, priced 75-90% lower than Haiku 4.5 for prompts under 100,000 tokens. The model introduces an adjustable effort setting and achieved 72.4% on the OSWorld benchmark, up from 15.7%, outperforming OpenAI's GPT-6 Luna (48.9%). Asana reported over 30% lower latency and up to 2.5x faster inference. Anthropic positions Haiku as a subagent under its Opus and Sonnet models.
Editorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
User reports 80% cost drop running transcripts through Anthropic's Haiku 5.5 vs 4.5
A user reported running 20 podcast transcripts through Anthropic's Haiku 5.5 model with the same prompt used for Haiku 4.5, resulting in an 80% reduction in their bill. The user noted that Anthropic had previously stated a 75% cost reduction. The user also observed that thinking mode is now enabled by default, causing output to grow 3.9 times, and that their old 1,000-token cap cut off 17 out of 20 answers mid-sentence. The report is based on one task type and includes no quality benchmark.
Anthropic cuts cheapest model price 10x, new Haiku beats OpenAI small model on computer tasks
Anthropic has reduced the price of its cheapest AI model by 10x, with the new Haiku 5.5 model outperforming OpenAI's small model on computer operation tasks, scoring 72.4% versus 48.9% on the OSWorld benchmark. Haiku 4.5 charged $1 per million input tokens and $5 per million output tokens, while Haiku 5.5 charges $0.10 and $0.50 respectively for prompts under 100,000 tokens, which covers 90% of the traffic the old model served. Cache reads dropped from $0.10 to $0.01. An always-on agent consuming a million tokens per hour costs approximately $2.40 per day in input. The author notes that while frontier models win benchmarks and headlines, small models win deployments because most production AI involves tasks like summaries, classification, and routing, where cost per task is critical. Anthropic positions Haiku as a subagent under its more expensive Opus and Sonnet models, with the expensive model planning and the cheap model executing hundreds of tasks in parallel, effectively making the labor at the bottom of the AI org chart 10x cheaper overnight.
Read sourceAnthropic's Haiku 5.5 AI model offers 100k token context at 10x lower cost than Haiku 4.5
In a post on X, user bcherny states that Haiku 5.5 is a good AI model, claiming it features a 100,000-token context window and is 10 times cheaper than its predecessor, Haiku 4.5. The post attributes these specifications and the cost comparison directly to the unnamed source's assessment. No further details about the model's capabilities, release date, or availability are provided. The claim suggests a significant improvement in both performance and affordability for the Haiku model line, which is part of Anthropic's AI model family. The post does not include any official confirmation or additional context from Anthropic.
Show 8 older updatesHide older updates
Anthropic releases Haiku 5.5 with 90% cost reduction and major benchmark gains
Anthropic has released Haiku 5.5, a new AI model that costs 90% less than Haiku 4.5 on prompts up to 100K tokens. The model introduces the first Haiku effort setting, allowing users to trade cost for accuracy. Despite these improvements, Anthropic still recommends its Sonnet 5.5 and Opus 5.5 models for complex agentic coding tasks such as Terminal-Bench 4.0. Asana reported over 30% lower task-completion latency and up to 2.5x faster inference per agent turn compared to its current model. On the OSWorld 2.1 benchmark, which tests AI agents operating real computers for long multi-step tasks, Haiku 5.5 achieved 72.4%, a significant jump from Haiku 4.5's 15.7%. On Chartography, a visual reasoning test for reading and interpreting charts without tools, the model rose from 6.4% to 46.4%.
Read sourceAnthropic's Haiku 5.5 delivers 3-5x benchmark improvements over Haiku 4.5, beats GPT-6 Luna
In a post on X, user ChrisGPT announced that Anthropic's Haiku 5.5 model represents an almost exponential jump over the previous Haiku generation, with several benchmarks improving by 3-5x over Haiku 4.5. Specific benchmark scores cited include OSWorld rising from 15.7% to 72.4%, HLE without tools from 10.2% to 45.9%, HLE with tools from 18.7% to 57.4%, and Chartography from 6.4% to 46.4%. The post claims Haiku 5.5 even beats GPT-6 Luna on computer use, knowledge work, FrontierCode, and visual reasoning. Pricing is listed as $0.10 per million input tokens and $0.50 per million output tokens for contexts under 100K, and $0.50/$2.50 for contexts over 100K. These results and comparisons are attributed to the source and have not been independently verified.
Read sourceAnthropic's Haiku 5.5 beats OpenAI's GPT-6 Luna on every benchmark compared
A post on X by user DanDr1s reports that Anthropic's AI model, Haiku 5.5, outperformed OpenAI's GPT-6 Luna on every benchmark that Anthropic compared. Specifically, Haiku 5.5 achieved a score of 72.4% on the Computer use benchmark versus GPT-6 Luna's 48.9%. On Terminal-Bench, Haiku 5.5 scored 39.2% compared to GPT-6 Luna's 16.4%. The post concludes that OpenAI's cheaper model has been surpassed.
Anthropic launches Haiku 5.5 AI model, claiming 10x cost reduction under 100k tokens
Anthropic has announced the release of Haiku 5.5, a new version of its AI model, claiming it is cheaper and faster than previous iterations. The company states that Haiku 5.5 is 10 times cheaper than Haiku 4.5 for usage under 100,000 tokens. The model is available for computer use, workflows, and through the API. The announcement follows user demand for the Haiku model line.
Read sourceAnthropic releases Claude Haiku 5.5, claiming 75% cost reduction over Haiku 4.5
Anthropic has released Claude Haiku 5.5, a new small AI model that is described as cheap, fast, and very capable. According to the announcement, the model is approximately 75% cheaper to run than its predecessor, Haiku 4.5, and features an adjustable effort setting. The poster, identified as addyosmani, states that their team loves using Haiku 5.5 as a subagent alongside the larger Opus 5.5 model. The post attributes the cost comparison and capability claims to the source, without independent verification.
Anthropic releases Claude Haiku 5.5, its fastest and cheapest model, costing 75% less to run
Anthropic has released Claude Haiku 5.5, described as its fastest and cheapest model to date. The new model costs 75% less to run than its predecessor, Haiku 4.5. It is the first Haiku model to feature an adjustable effort setting. According to the announcement, the model shows a significant leap in performance on benchmarks for Computer Use and Agentic Coding.
Read sourceAnthropic launches Claude Haiku 5.5 with 75% lower running costs than Haiku 4.5
Anthropic has launched Claude Haiku 5.5, a new AI model that the company claims offers average running costs approximately 75% lower than its predecessor, Haiku 4.5. The post, attributed to user kimmonismus on X, states that the model is 'way better than GPT-6-Luna' and 'close(r) to Sonnet 5.5' in performance. Anthropic says the model also brings improvements in coding, computer use, and knowledge work. Additionally, Haiku now features adjustable effort, allowing developers to choose how much intelligence to allocate per task. The post expresses strong approval, calling it a 'super good release' that is both cheap and smart. These claims are attributed to Anthropic and the poster's opinion, with no independent verification provided.
Read sourceAnthropic releases Haiku 5.5 with pricing at $0.10/M input tokens
Anthropic has released Haiku 5.5, a new AI model, with pricing starting at $0.10 per million input tokens and $0.50 per million output tokens for usage under 100,000 tokens. For usage over 100,000 tokens, the price increases to $0.50 per million input tokens and $2.50 per million output tokens. The announcement, made via a post on X by user Mr_Salio, characterizes the pricing as very cheap and suggests Anthropic has performed well with this release. The post includes a link to further details.
Read source