Haiku 5.5 Brings Lower-Cost AI Teamwork
Haiku 5.5 is available in Claude Platform and Claude Code, with around 75% lower average run cost than Haiku 4.5 and a supporting role alongside Opus or Sonnet.
THE LOG — TRANSMISSIONS FROM ALTITUDE
What we learn running AI systems every day — workflows, growth loops, tools, and the occasional hard lesson. Written down so you can reuse it.
Haiku 5.5 is available in Claude Platform and Claude Code, with around 75% lower average run cost than Haiku 4.5 and a supporting role alongside Opus or Sonnet.
Interactive answers, in-file edits and typed decisions bring AI closer to execution. The review boundary matters as much as the interface.
OpenRouter lists a huge context window, text-and-image inputs and tool support for Mistral Large 4. Here is what that offers—and what still needs testing.
The 7 October daily briefing: AI moves closer to editing files, configuring agents and choosing actions, with different controls for human review.
Claude says it can read and edit our open Google Workspace file from a sidebar, with an opportunity to approve each edit before it lands.
Today's AI briefing separates output speed, chosen-voice listening and output warnings from evidence of better workflow outcomes, alongside OpenClaw's release repair and Aleph Alpha's Kolibri.
Persistent agents, new models and inspection tools put the gap between availability and safe deployment at the centre of this week's AI briefing.
OpenAI claims faster token generation with Ultrafast in Codex and the API. Here is what those figures mean—and what they do not establish.
ClaudeDevs says Claude Code's You should Know plugin scans output for important information we might miss. Its demo illustrates a consequence worth reviewing, not a safety guarantee.
ElevenLabs says Eleven v4 is now in ElevenReader, bringing chosen-voice listening to articles, ebooks and PDFs. The announcement is not a listening test.
DeepSeek announces packaged Harness desktop versions for macOS and Windows and a separate Linux npm route. The captured announcement establishes access routes, not tested capabilities.
Decision models, typed choices and tutoring evaluations put measurement at the point of action, alongside updates to coding agents, image generation and voice tools.
New models promise faster, cheaper work, but buyers still need to test quality, latency, retries and total cost on their own tasks.
Reports of AI agents reaching US government websites put attribution under scrutiny. A separate Muse report raises a model-lineage question, while decision-model packaging and search tools spread. We examine what operators can verify—and what remains open.
This week’s releases point beyond a contest of model names. Speech, images, video and browser-side interaction are being packaged into tools people can use inside existing workflows.
Microsoft’s OpenClaw-based Autopilot signals a move from agent demos toward adoption, while disclosures and a court ruling show how scrutiny is arriving through very different channels.
Managed surfaces are changing how agents gain capabilities
Where does your AI agent run—and how do you reach it?
Google, ElevenLabs and OpenAI are building different layers of the voice stack. Which work is easier now, and what still needs production proof?
Muse Realtime Avatar promises a responsive, consistent character. Meta’s demos show the ambition, but not broad app availability or independently verified superiority.