Igor's KB Site

A personal website for information sharing

Follow me on GitHub

AI and LLM Weekly — 31 October 2026

· ai · 6 items

Google's Gemini 4 Carbon tops an intelligence index with cyber-first access, while Anthropic details exploit chains from an unreleased model; six items this week.

Google unveils Gemini 4 Carbon, its strongest model, starting with cyber defenders

Google · 28 Oct 2026 · Model release

Carbon arrives about two weeks after Gemini 4 Argon and is offered first to a subset of cyber defenders, then to trusted partners through the Fairwind Program. Google reports a score of 59 on the Artificial Analysis Intelligence Index, 88.2% on SEC-bench Pro for finding vulnerabilities and 71.0% on CWE-bench for patching, all tested without guardrails. Introductory pricing matches Argon at $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 later. These are company-reported figures, and no date is given for general availability.

Anthropic describes five attack chains built by an unreleased research model

Anthropic · 27 Oct 2026 · Security

The Frontier Red Team says an internal model, with some prompting for each chain, found bugs and assembled working exploits against Chrome, Safari, a locked Pixel 9, Windows HTTP/3 and the Linux multipath TCP stack. The Chrome work alone produced 62 critical or high CVEs, fixed in Chrome 156. Anthropic widened its Cyber Verification Program and published a Claude Code security-guidance plugin, but the model itself stays unreleased. The account is Anthropic’s own and has not been independently verified.

Mistral publishes open weights for Large 4, a 1-trillion-parameter multimodal model

Mistral AI · 29 Oct 2026 · Open weights

This follows the public preview covered earlier this month, and the release is an improved version of that model. The license is a modified MIT that allows commercial use, except that firms offering the model as a hosted service need a separate commercial license. Mistral says Terminal-Bench 4.0 rose 13 points from the preview to 41%, the best among open-weight models, and lists API pricing of $0.70 per million input tokens and $2.10 per million output. The page is undated, so the date comes from press coverage, and the technical report is still to come.

OpenAI ships GPT-6.1 Terra, a cheaper mid-tier model for computer use

OpenAI · 26 Oct 2026 · Model release

Terra takes text and image input, offers a 1.05-million-token context window, and costs $1 per million input tokens and $5 per million output. Its documentation positions it for visual reasoning, computer use and drafting, with reasoning effort from none to max and tool calling through the Responses API. Fine-tuning is not supported. Press coverage reports that OpenAI still points users to larger models for deep research and complex coding. This links to the model documentation because OpenAI’s announcement page could not be retrieved.

DeepMind announces Genie 4, a real-time interactive world model with native audio

Google DeepMind · 27 Oct 2026 · World models

Users move through generated scenes with keyboard controls and steer them with text prompts, changing weather, camera or objects as the simulation runs. DeepMind names robot and agent training environments among the uses and says outputs carry SynthID watermarks. It is a research preview for trusted testers, with a waitlist and planned access via Google AI Ultra and the Gemini API. DeepMind concedes that object permanence and visual detail drift over longer sessions.

Anthropic reports early results from Claude-assisted drug development

Anthropic · 27 Oct 2026 · Science

Working with six partners, Anthropic describes a shortened CAR T genetic component that raised receptor production about tenfold in lab tests, and a document-review task at Enveda that cut human review time by 89%. Anthropic has also started its own programs, including one for rare muscular dystrophies. Nearly all of this is preclinical, none of the drugs has reached human testing, and the results are self-reported by a company with a stake in them.


← All digests · Site home