Claude Haiku 5.5 costs less; free ChatGPT users start getting GPT-6

Marcus here.

The price of AI work now depends on who gets the job.

Anthropic's Claude Haiku 5.5 costs 90% less than Haiku 4.5 for shorter requests, by Anthropic's count, which makes the small-model end of the bill worth another look.

OpenAI brings GPT-6 to Free and Go users today, extending yesterday's paid rollout across a service that OpenAI says reaches over 1.2 billion people weekly.

Elon Musk says Grok Bot will hand some tasks to outside models, Claude Opus 5.5 among them. Anthropic already has an agreement for all the compute at SpaceX's Colossus 1, so the guest has rather comfortable accommodations.

Stay curious,

Marcus Schuler

Anthropic released Claude Haiku 5.5 on October 7, adding adjustable effort settings to its model for high-volume, cost-sensitive work.

Five effort settings let users tune work such as summarization. Per million tokens, requests up to 100,000 tokens cost $0.10 input and $0.50 output, versus Haiku 4.5's $1 and $5. Above that threshold, prices rise to $0.50 and $2.50.

Claude Haiku 5.5 is available through Claude Platform and Amazon Web Services, with Google Cloud and Microsoft Azure also offering it. Anthropic halved Claude Sonnet 5.5 cache-read pricing to $0.10 per million tokens.

Why This Matters:

• Teams processing shorter requests pay less for routine summarization, with Claude Haiku 5.5's lower prices also applying to classification.
• Adjustable effort gives developers another cost control to test before assigning more customer work to Claude Haiku 5.5.

OpenAI began GPT-6's global rollout to ChatGPT Plus, Pro, Business and Enterprise on October 7, with Free and Go access starting October 8.

Paid users get GPT-6 Sol; Free and Go get GPT-6 Luna, with Intelligent UI adding charts, buttons and small tools to answers that can start while ChatGPT is still thinking. An administrator must enable it for Enterprise workspaces, and Work and Codex models stay unchanged. A savings calculator can now sit inside the conversation.

Scientific American examines how OpenAI's release of 372 mathematical results leaves mathematicians struggling to assess their novelty.

OpenAI says nearly all came from one prompt to one agent, although some may have required multiple attempts. With the specific prompts withheld, mathematicians have less information to judge which results are new, following an earlier Navier-Stokes effort involving 10,000 agents and millions in costs.

• Healthleap raised $38 million for software that reviews adult inpatient records nightly and flags malnutrition or delirium risks for clinicians to assess.
• Meta and Sierra announced Personal Agent Protocol, an OAuth-based standard governing how personal AI agents sign in to businesses and what they may do.
• Google faces an October 23 construction halt at two Finnish data center sites pending environmental assessments, affecting its $15 billion Finland plan.
• Google SynthID Detector opened to everyone to check AI-made images, video and audio, with people already making about 1 million verification requests daily.
• Microsoft opened U.S. preorders for the $5,999 Surface RTX Spark Dev Box, featuring Nvidia's RTX Spark chip and 128GB of unified memory, shipping in November.

A KPI can improve on paper because its denominator or time window changed. Use this before approving an operating review.

Your raw input:

Paste the current and prior KPI claims, both definitions, reporting dates, footnotes, source notes, and the decision the number will support.

The prompt:

Why this works: The model must inspect how the metric is built before interpreting movement. The fixed fields expose changed populations, periods, or exclusions that a polished slide can hide.

What to use: Use a reasoning model with document analysis. For sensitive figures, use a company-approved enterprise model or a local model.

NousResearch/hermes-agent is an open-source AI agent under the MIT license, with about 252,000 GitHub stars as of October 7.

What it does: Hermes keeps memory across sessions and searches past conversations. After complex tasks, it writes skills for itself and improves them through use.

Who needs it: People running unattended jobs can use its scheduler from a terminal or messaging gateway, choosing their model provider with hermes model.

Nous Research announced a $90 million Series B led by Robot Ventures on October 7 to build Hermes for Businesses. Hermes Agent on GitHub →

Elon Musk announced on October 7 that Grok Bot will choose outside AI providers by task, including Anthropic's Claude Opus 5.5. The announcement does not say Grok's own models have been abandoned.

Our take: Grok Bot's provider menu is an awkward advertisement for Musk's own models. Customers can get Claude Opus 5.5 through the bot, depending on the task. Anthropic already has a May 6 agreement to use all the compute capacity at SpaceX's Colossus 1. The rival has booked the building and now gets a place on the menu.

That compute agreement sets no terms for Grok Bot routing. Grok's security guidance tells customers whose contracts restrict outside data processors to talk to their account team before deploying. Musk's bot can keep working after the laptop closes. The customer may want to keep the contract open.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai