Claude Haiku 5.5 Is Here
Anthropic launched Claude Haiku 5.5, its fastest and most affordable model yet. It costs $0.10 per million input tokens and $0.50 per million output tokens, roughly 75% cheaper than Haiku 4.5.
Anthropic says it beats Haiku 4.5 by a wide margin, approaches Sonnet 5.5 and tops GPT-6 Luna on the benchmarks it shared. It is on the Claude Platform, AWS, Azure and Google Cloud.
Cyber Verification Program Splits Into Three Tiers
Defense Access covers incident response and malware analysis. Red Team Access adds authorized pen-testing. Specialized Access has the fewest safeguards and is limited to verified organizations. Tiers cover Opus 5.5, Sonnet 5.5 and Mythos 5.1.
Project Glasswing has found 129,000+ verified vulnerabilities, over 33,000 rated critical or high. VulnCheck found only 0.67% of a 300-flaw sample exploited in the wild.
Claude for Google Workspace Hits Public Beta
Add-ons for Docs, Sheets and Slides are now in public beta on all paid plans. New beta connectors let Claude create and edit Google files directly.
Credits for Max and Team, Cheaper Caching
Max and Team subscribers now get monthly Claude Platform credits: $100 on Max 5x, $200 on Max 20x, and up to $500 pooled on Team. Sonnet 5.5 prompt cache reads were cut in half to $0.10 per million tokens.
Heads up: Haiku 5.5 uses adaptive thinking, so code sending manual budget_tokens will get a 400 error.
SDKs, Managed Agents and Claude Code
The Python and TypeScript SDKs add beta classes for browser use and computer use tools. In Managed Agents, allowed_hosts under limited networking now also restricts web_search and web_fetch. The Models API reports per-model capability flags.
Claude Code 2.1.293 makes Haiku 5.5 the default Haiku model. Earlier 2.1.29x builds added an effort parameter on the Agent tool and a --marketplace option for plugin installs. Enterprise Compliance API chat endpoints now include unified Claude chats (beta).
Pentagon Says It Stopped Using Claude
BBC reports the Pentagon says it has stopped using Claude, while BBC sources say the model was used in Iran operations the previous week. Details remain thin.
Meta and Microsoft Reportedly Steer Staff Elsewhere
The Information reports Meta cut internal Claude Code users to about 30K as Meta and Microsoft push employees away from Anthropic tools.
Ten Cents Changes the Default
Haiku 5.5 at $0.10 per million input tokens, with 1M context and a cheaper cache on Sonnet, makes sub-agents and bulk jobs far less painful. Expect it to become the default worker model.
Bundled API credits point the same way: Anthropic wants subscribers building on the platform, not just chatting. SemiAnalysis argues Claude's $200 plan already delivers about 5x the API-equivalent value of OpenAI's mid-tier plans.