Anthropic released Claude Haiku 5.5 and cut Sonnet cache-read pricing

TOP 3

Models

Claude Haiku 5.5: In its October 7 launch, Anthropic positions the new small model for high-volume tasks, including summaries, classification, browser use and coding subagents. It is available through the Claude Platform as claude-haiku-5-5 and on AWS, Google Cloud and Azure; it is the first Haiku-class model with adjustable effort. Anthropic lists input at $0.10 and output at $0.50 per million tokens for prompts up to 100,000 tokens, with higher rates above that threshold. It says average running cost is around 75% below Haiku 4.5, accounting for task token usage; this is not a flat 75% discount on every request.
- The accompanying change and the limit: Anthropic also halved Sonnet 5.5 cache reads to $0.10 per million tokens, from $0.20, and says this makes most agentic work around 20% cheaper. Its benchmark comparisons and safety results are Anthropic’s evaluations, not independent verification here. The company itself still recommends Sonnet or Opus over Haiku for complex agentic coding; the promised monthly API credits for Max and Team subscribers roll out later this week, rather than being part of the model’s immediate availability.
Open d1 decision models: Liquid AI’s October 7 release makes d1-3B (text and image) and experimental d1-omni-600M (text and image or text and audio) weights available. These models answer structured questions in a forward pass rather than generating tokens; the speed and benchmark figures are Liquid AI’s own, and it does not report vision or audio benchmark results.

Agentic AI & Agent Skills

Deep Agents skills: LangChain’s October 7 update adds tools bound to skills that load when the skill is read, runtime-pinned skills that load before the next model call, and a way to refresh the skill library mid-thread; prompt-cache preservation for newly added tools depends on the model’s support.
Managed Deep Agents: LangChain’s October 7 v0.9 release adds agent-created schedules, per-run selection of model, skills and tools, and Slack acknowledgement reactions. It remains in public beta; scheduled runs use the requester’s permissions and return to the originating channel.

Tools & Open Source

Windows local AI: NVIDIA’s October 7 account of its Microsoft event says RTX Spark laptop preorders opened, with devices due October 16 and compact desktops in November. Microsoft announced general availability of Microsoft Execution Containers for agents running under OS control; NVIDIA also previewed, rather than released, DGX Station for Windows. The stated RTX Spark ceiling is 128GB of unified memory.
Prompt-to-game creation: Google’s October 7 announcement introduces Playground, a browser-based experimental platform to create and share games through prompts. It launches for U.S. adults with tiered creation access by Google AI subscription; the Unity Spark integration is still in testing, not a launch feature.