K Koda Intelligence
Subscribe
Cover image for the 29 October 2026 Signal: Scoring 59 on the Intelligence Index, one point above Claude Opus 5.5, Gemini 4 Carbon reaches cybersecurity defenders first.
THE SIGNAL · 29 OCT 2026 · 9 MIN READ

Gemini 4 Carbon goes to defenders first

Google announced Gemini 4 Carbon on October 28. It scored 59 on the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 at 58, and Google has set no date for wider release.

KI Koda Editorial · Updated 02:12 GMT · 26 sources
AICYBERSECURITYMODELS
33 of 46 figures cited19 receipts in the register
  1. 88.2%1Ars Technica
  2. 71.0%2Ars Technica
  3. 85.6%3CNN
The register
Lead Story

The day's defining move.

Model Release · Ars Technica
Lead storyModel Release 88.2%1 Ars Technica · receipt 01, verified
Model Release·Ars Technica·28 OCT 2026

Gemini 4 Carbon Launches for Cyber Defenders First

Google announced Gemini 4 Carbon on October 28. It scored 59 on the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57. Initial access is limited to a group of cybersecurity defenders, and Google has given no firm date for developer, enterprise or consumer availability.

Why it mattersUnless you are among the cybersecurity defenders with early access, you cannot build on Gemini 4 Carbon yet.

Continue reading
World

With 85.6%3 of ballots counted after Tuesday's election, opposition parties held 67 of 120 Knesset seats, leaving Netanyahu's Likud short of a governing majority.

Markets

The S&P 500 rose 0.28% and WTI crude gained 1.91%, while bitcoin edged up 0.02% and ether slipped 0.15%; that split left crypto without a clear direction, matching the crypto Fear & Greed index's Neutral reading of 51.

SPONSOR KODAFrom the desk

Put your product in front of AI operators.

Koda publishes a fact-checked AI intelligence briefing every morning: site, email, podcast, and video. Founding sponsors lock today's rates for 6 months as the audience compounds.

View the media kit →
Listen & Watch

Daily broadcasts.

Podcast · Video · Infographic
YouTube · Short

Today's Signal Short

Visual

Intelligence map

Markets

6 levels at the close.

Market TerminalLAST CLOSE 28 OCT 2026 · READ 2026-10-29 · Yahoo Finance
S&P 500 7,743.9813 +0.28%13 6D 7,623.68 low / 7,757.17 high
Nasdaq 26,549.8514 +0.27%14 6D 25,955.34 low / 26,702.86 high
Bitcoin $80,823.4315 +0.02%15 6D $80,006.66 low / $81,074.19 high
Ethereum $2,344.1216 -0.15%16 6D $2,320.28 low / $2,384.08 high
Crude Oil (WTI) $101.8117 +1.91%17 6D $97.15 low / $103.27 high
Crypto Fear & Greed 51 / 10018 Neutral 0 to 100 index 25 fear / 50 neutral / 75 greed · alternative.me
Today's Focus

What else the day turns on.

01

Mizuho Cuts Circle Target

Mizuho cut its Circle price target to $43 from $45 and kept an Underperform rating, implying about 46% downside from Monday's $79.84 close. Its simulation has AI personal assistants steering USDC holders toward higher-yielding FDIC-insured options such as SoFi savings, which could cut Circle's USDC reserve revenue by up to 20%4.

Source: tldr.tech
02

Figma Weave Inside External Agents

Figma Weave's video and image models, including Seedance, Kling and Nano Banana, are now usable inside external AI agents via the Figma MCP. That moves Weave's generation tools out of Figma's own interface and into third-party agent workflows.

Deep Dive

Today's editorial.

All editorials
Deep Dive cover image: Google Gave Defenders Its Best Model First
Deep Dive6 min read

Google Gave Defenders Its Best Model First

Read full analysis
The Wire

7 stories from 6 desks.

Lead carried above
Open Source·VentureBeat·27 OCT 2026

Liquid AI Ships LFM3 Small Models for Devices

Liquid AI released LFM3 on October 27, a family of roughly 2-billion-parameter text, vision, audio, omni and decision models.

Read
Hardware·Reuters·26 OCT 2026

Nvidia to Invest $3 Billion5 in Meta's Offering

Nvidia will buy into the equity offering Meta announced this week to fund AI, as the two expand their partnership. Its commitment is $3 billion5, according to TLDR.

Why it mattersHold any read of Meta's offering as broad market demand until the final allocation shows how much outside investors took beyond Nvidia's $3 billion5.

Read
Enterprise·OpenAI

OpenAI Expands ServiceNow Partnership Across Enterprise Workflows

ServiceNow is OpenAI's latest expanded enterprise tie-up, aimed at software development and security across platforms, arriving the same week Oracle apps came to ChatGPT.

Why it mattersServiceNow admins should ask which OpenAI models and data terms the expansion covers, the same questions Oracle buyers now face with Oracle apps inside ChatGPT.

Read
Also on the wire
  1. AnthropicAnthropic Expands AI for Science to 20 Institutions
  2. Gainesville Sun (press release)Riemann's 13.1B6 English-Arabic Model Passes 500 Downloads
  3. OpenAIOpenAI Adds ChatGPT Plugins for Apple Mail, Calendar
The Globe

3 stories off the AI beat.

Economy·Reuters·28 OCT 2026

Fed Pauses Rate Hikes, Holds Benchmark Steady

After raising rates last month, the Federal Reserve held its benchmark at 3.75%7 to 4.00%8 on Wednesday while officials assess persistent inflation.

Why it mattersBorrowers financing data center and chip projects get no relief yet, since the Federal Reserve's pause amid persistent inflation leaves borrowing costs where last month's hike put them.

Read
Also on the globe
  1. ReutersUS Iran War Costs Top $47 Billion9 Mark
  2. CNNIsrael's Opposition Wins 67 of 120 Seats
The Lab

Tool of the day, researched.

All Lab reports
Droidtty home page, captured for the 29 October 2026 Lab report Tool of the day 6.519/ 10
Coding Free

Droidtty

An agent run that stalls on one permission prompt can sit idle until morning, and Droidtty's whole pitch is getting that prompt onto your Android phone first.

Droidtty does one job the general mobile terminals skip. It reads your agent's state and turns a stalled permission prompt into a decision card you answer with a tap. Adopting it means a sideloaded beta plus a desktop bridge on your own network, and nobody outside the project has published a hands-on review, so trust it on a short run before an overnight one. Developers juggling multi-hour Herdr jobs should install it now; iPhone users can wait.

Capability 6.5
Ease 6.0
Value 8.5
Momentum 4.5
Read the Lab report Try it
Also on the radar
Build

Anything: turn a plain-language idea into a working app

Anything is a no-code app builder that converts a written description into a functioning app, with GPT-5 models built into the workflow. Start with one narrow feature, such as a booking form or an internal tracker, and test it with real users before you add scope. There is a free tier, and paid plans start from $1612/month if you outgrow it.

Try it
Productivity

NeuralTalk AI: build a custom business chatbot without code

NeuralTalk AI lets businesses create their own conversational chatbots without writing code. Begin by loading your top 20 customer questions and their approved answers, then run the bot against real support tickets to find gaps before you put it on your site. Keep a human handoff path for anything involving refunds or account changes.

Try it
Build

OpenBot: self-hosted agents that each get their own machine

OpenBot is a forkable, self-hosted agent platform that gives every bot a dedicated machine, so one agent's files and processes stay isolated from the others. Use it for private automations you don't want running on a shared cloud service, and fork an existing bot as a template when you build a new one. Isolation also limits the damage if a single agent misbehaves.

Try it
Productivity

SquadRoom: specialist agents that deliver a finished report

SquadRoom is a workspace where several specialist AI agents collaborate on a board-level task and hand back a completed report. Give it a tightly scoped brief, such as a competitor comparison or a quarterly market summary, and name the sections you expect so the output arrives in a usable shape. Check the sources behind any figure before it reaches a decision-maker.

Try it
The Arena

1 lab moved this week.

  1. AnthropicAnthropic expanded its AI for Science program to 20 scientific foundations, institutions and research groups worldwide. anthropic.com · 28 OCT 202601
Builder Radar

One paper, read at the source.

Primary sources, pulled by index
Paper of the day

Kolibri: A Sovereign European Model on the Pareto Frontier

Aleph Alpha, Moritz Althaus, Artur Baranowski et al. · arXiv 2026-10-22

This report introduces Kolibri, an English-German mixture-of-experts transformer with 78.1B total parameters and 3.46B active parameters per token, released as open weights under the Apache 2.0 licence.

This report introduces Kolibri, an English-German mixture-of-experts transformer with 78.1B total parameters and 3.46B active parameters per token, released as open weights under the Apache 2.0 licence.
Read the paper

The paper comes from the Firecrawl Research Index: arXiv, indexed in the last 7 days and submitted this month or last, picked for today's lead. Quotes are verbatim.

THE DOJO · BUILD TODAY

Shorten the distance between your agents and your decisions.

  1. 01

    Map your agent's approval stalls. List every point where a coding agent waits on a permission prompt, then note how long each sits unanswered. That list tells you whether a tap-to-answer tool like Droidtty is worth a sideloaded beta.

  2. 02

    Run a find-versus-fix check on your own code. Use a model you can reach today to list flaws in one service, then time how long it takes to patch them. Carbon's reported 88.2% finding score against 71.0% fixing gives you a yardstick for what to measure.

  3. 03

    Wire Figma Weave into an agent. Weave's image and video models, including Seedance, Kling and Nano Banana, now work inside external agents through the Figma MCP. Prototype one asset-generation step and note where the agent hands control back to you.

The Bottom Line

Defenders get the best model first, and builders wait.

Google reports Gemini 4 Carbon scores 88.2%1 on SEC-bench Pro, which tests finding flaws, and 71.0%2 on CWE-bench, which tests fixing them. Its exploit skill grew fastest, and API access remains gated. While you wait for access, a tool like Droidtty shows how to tighten the approval step of your own agent workflows, so a gated model has somewhere to land when it opens up. Start with the stalls you can measure today.

Source: arstechnica.com
Receipts

Every number, and where it came from.

33 of 46 figures cited, 19 receipts: 4 verified, 12 reported, 3 estimated
  1. 0188.2%Gemini 4 Carbon Scores 88.2% on Vulnerability Detection.Ars Technica28 OCT 202603B search corroborationV
  2. 0271.0%...multi-day agent work, scoring 88.2% on SEC-bench Pro and 71.0% on CWE-bench patching.Ars Technica28 OCT 2026as publishedR
  3. 0385.6%With 85.6% of ballots counted, opposition parties held 67 of 120 Knesset seats, leaving Netanyahu's Likud short of a...CNN28 OCT 2026as publishedR
  4. 0420%Mizuho: AI agents could cut Circle's USDC reserve revenue by up to 20%Mizuho, via TLDR28 OCT 2026as publishedE
  5. 05$3 BillionNvidia to Invest $3 Billion in Meta's Offering.Reuters26 OCT 202603B search corroborationV
  6. 0613.1BRiemann's 13.1B English-Arabic Model Passes 500 Downloads.Gainesville Sun (press release)28 OCT 2026as publishedR
  7. 073.75%...rates last month, the Federal Reserve held its benchmark at 3.75% to 4.00% on Wednesday while officials assess...Reuters28 OCT 202603B search corroborationV
  8. 084.00%...month, the Federal Reserve held its benchmark at 3.75% to 4.00% on Wednesday while officials assess persistent...Reuters28 OCT 202603B search corroborationV
  9. 09$47 BillionUS Iran War Costs Top $47 Billion Mark.Reuters28 OCT 2026as publishedR
  10. 101.7BLiquid AI launches LFM3, open weight on-device models whose 1.7B text model beats its 2.6B predecessor and fits in...Liquid AI, via TLDR28 OCT 2026as publishedR
  11. 112.6B...open weight on-device models whose 1.7B text model beats its 2.6B predecessor and fits in about 1 GBLiquid AI, via TLDR28 OCT 2026as publishedR
  12. 12$16There is a free tier, and paid plans start from $16/month if you outgrow it.theresanaiforthat.com29 OCT 202603B search corroborationE
  13. 137,743.98S&P 500 close, +0.28% on the sessionYahoo Finance28 OCT 2026yfinance closeR
  14. 1426,549.85Nasdaq close, +0.27% on the sessionYahoo Finance28 OCT 2026yfinance closeR
  15. 15$80,823.43Bitcoin close, +0.02% on the sessionYahoo Finance29 OCT 2026yfinance closeR
  16. 16$2,344.12Ethereum close, -0.15% on the sessionYahoo Finance29 OCT 2026yfinance closeR
  17. 17$101.81Crude Oil (WTI) close, +1.91% on the sessionYahoo Finance28 OCT 2026yfinance closeR
  18. 1851 / 100Crypto Fear & Greed close, Neutral on the sessionalternative.me29 OCT 2026as publishedR
  19. 196.5 / 10Droidtty, Koda score across four dimensions04R dossier29 OCT 202604R dossier scoringE
V
Verified: the stat gate corroborated this figure against an independent search before publication.
R
Reported: the figure is carried as published by the linked source and was not independently corroborated.
E
Estimated: the figure is approximate or hedged, either in the source or by the stat gate.

Method names the computation path behind the row, from a closed vocabulary: as published, yfinance close, 03B search corroboration, 04R dossier scoring, not recorded.

A figure is money, a percentage, a spelled magnitude, a multiple, a rate carrying its unit, a score over its denominator, a thousands separated number or an index close. Bare integers and years are not counted, so a phrase like "over 100 integrations" carries no mark. A chart's own scale line is the axis of the level above it and is counted once, with that level.

Get the morning Signal

215 editions so far, one a day. Unsubscribe anytime.

Forward this to one operator you work with. Your referral link is in every email; milestones at koda.community/refer.