Claude Haiku 4.5
Claude Haiku 4.5 is a model in Anthropic’s Haiku line, released on October 15, 2025 (API ID claude-haiku-4-5-20251001). It takes text and image input, returns text, and supports extended thinking.
Superseded. Anthropic lists Claude Haiku 4.5 as a legacy model that is still available; its current Haiku model is Claude Haiku 5.5.
Released
October 15, 2025
Type
multimodal
License
proprietary
Context
200,000 tokens
Up to 64,000 output tokens.
Pricing
Input
$1.00 / 1M tokens
Output
$5.00 / 1M tokens
List price on the developer's own pages, last checked , more than 7 days ago: it may have changed since.
Benchmarks
| SWE-bench VerifiedAnthropic’s figure: all 500 tasks, average of 50 trials, 128K thinking budget. | 73.3% |
Scores are the developer's own published figures for this model, under the test conditions noted. Benchmarks with no primary source are not listed.
Capabilities
Sources
Every figure on this page was checked against these pages on . A figure we could not source is left out.
- Anthropic news: Claude Haiku 4.5 (opens in a new tab)
- Claude docs: Claude Haiku 4.5 (opens in a new tab)
- Claude docs: Models overview (opens in a new tab)
Spotted something out of date? Tell us.
Links
In the news: Claude Haiku 4.5 and Anthropic
The Hacker News
Linux Kernel TCP Flaw With 9.8 CVSS Score Affects Ubuntu, Red Hat and Other Distributions
A critical vulnerability in the Linux kernel's TCP code, tracked as CVE-2026-98432, has been identified and affects distributions including Ubuntu and Red Hat Enterprise Linux (RHEL). The Linux kernel's own CVE team has assigned this flaw a severity score of 9.8 out of 10 on the CVSS scale. Anthropic reported on October 27 that its AI model successfully generated an exploit for this vulnerability. The exploit allows a remote attacker, without needing any credentials, to execute commands with root privileges on affected Linux servers.
The Hacker News
Fake ChatGPT and Claude Apps Abuse Signed Executables to Deploy Crypto Miners
Threat actors are distributing malicious loaders and cryptocurrency miners disguised as artificial intelligence (AI) desktop applications for Anthropic Claude and OpenAI ChatGPT. This campaign has been dubbed Yoba. The activity involves the creation of fake GitHub projects that falsely claim to be desktop applications such as GPT-5.6 Sol, GPT-6 Astra, and Claude Opus. These fake applications are used to deliver a malware family. These malicious applications abuse signed executables to deploy cryptocurrency miners, posing a security risk to users seeking AI tools.
MarkTechPost
Nace AI Releases Drex for Search for AI Agents: 92.05% on BrowseComp-Plus With 12.6 Search Calls
Nace AI has released Drex for Search, a retrieval layer that integrates its Drex decision models into document search to assist AI agents in finding evidence. This system is designed to be called by coding agents such as Claude Code, Codex, Cursor, and OpenCode via CLI, SDKs, MCP, or REST. In Nace's own evaluations, Drex for Search achieved a score of 92.05% on the BrowseComp-Plus benchmark when paired with GPT-5, utilizing an average of 12.6 search calls per question. The system is described as a hosted search solution capable of searching over one million files and can be paired with answering models like GPT-5, Gemini 3.5 Flash, or Qwen 27B. Drex for Search can be deployed on Nace's managed cloud, dedicated clouds, or users' own AWS or Azure accounts, as well as on-premises and in air-gapped environments. Nace reports that the system ranks first or second on three out of four public leaderboards, including achieving 77.5% R@1 on VisDocAgentBench, which is 10 points higher than Claude Opus 5. Nace AI frames search as a series of decisions, and Drex for Search applies its models to these decisions by scoring options with a calibrated probability in a single forward pass, making it cost-effective for each agent step. The system targets limitations in classic RAG pipelines, such as answers being located in tables, requiring cross-file hops, or being within scanned PDFs. Within an agent, Drex for Search operates in three steps: Classify, where it determines if a question requires a quick lookup or an agent route; Squeeze, where it narrows down the search space by navigating through folders, files, sections, and tables; and Answer, where only relevant pages and tables are sent to the LLM, with citations to the file and page number.
The A.V. Club
Anthropic sued for allegedly stealing from The King
Anthropic is being sued by the rights holders of Elvis Presley's music catalog, Authentic Brands Group, for allegedly using copyrighted songs to train its Claude large language model (LLM). The lawsuit claims Claude scooped up approximately 274 copyrighted Presley songs to generate AI-generated content. According to the filing, Claude has created infringing output of copyrighted material from virtually every song available on the internet, including the Elvis Presley works. Even after Anthropic implemented 'guardrails,' the lawsuit alleges that Claude continues to infringe. When recently asked to recreate specific songs for a modern audience, Claude reportedly admitted that its renders were very close to the originals, featuring the same structure, reused hooks with slight rewording, and full sets of verses, choruses, and bridges. Authentic Brands Group is seeking $150,000 per infringed Elvis song, totaling around $41.5 million. This lawsuit highlights the growing legal challenges surrounding AI's habit of scraping copyrighted material for training purposes.
MarkTechPost
Anthropic Launches OSS Scanner, a Free AI Vulnerability Scanner
Anthropic has launched OSS Scanner, a free, opt-in vulnerability scanner designed for critical open-source projects. This tool utilizes Anthropic's most powerful AI models, including Claude Mythos, to analyze enrolled codebases. Reports generated by OSS Scanner are sent directly to project maintainers without an initial human review. This approach aims to accelerate the discovery of vulnerabilities, though it shifts the triage work to the project teams. As of October 19, Anthropic reported 77 critical and 803 high model-rated findings. The program has accepted 114 projects out of approximately 650 applications. OSS Scanner runs on Anthropic's infrastructure at no cost to users. Maintainers provide a Dockerfile, and audits are conducted within an isolated virtual machine that has no internet access. Of the 97 audited critical and high findings, 85% met Anthropic's disclosure criteria. The scanner demonstrated a low false positive rate, with only 1 of 97 audited findings being a false positive. For example, wolfSSL stated that 72 out of 74 reports were valid. The design of OSS Scanner draws inspiration from Google's OSS-Fuzz, substituting fuzzers with language models to address review bottlenecks. Over a six-month period, Anthropic's models identified over 29,000 candidate vulnerabilities, with human reviewers triaging about 6,000. The OSS Scanner program is an evolution of an ad hoc request system where nearly 5,000 additional reports were sent directly to maintainers who had requested all findings, regardless of validation. To enroll, core maintainers submit a pull request to anthropics/oss-scanner, adding a project configuration file. This file requires repository details, a primary contact, and a Dockerfile path, with optional fields for threat models and encrypted reports.