K Koda Intelligence
Subscribe
Cover image for the 01 October 2026 Signal: Google's largest Gemini model is here, but only vetted cybersecurity partners in its Fairwind program can use it.
THE SIGNAL · 01 OCT 2026 · 9 MIN READ

Google limits Gemini 4 Argon to cyber partners

Google's largest Gemini model leads or ties on 14 of 19 evaluations and scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra. Only vetted Fairwind partners can touch it.

KI Koda Editorial · Updated 01:29 GMT · 25 sources
AIAGENTSSECURITY
22 of 23 figures cited13 receipts in the register
  1. 4,0001OpenAI, via TLDR
  2. 77.9%2Venturebeat
  3. $54 billion3Reuters
The register
Lead Story

The day's defining move.

Model Release · Ars Technica
Lead storyModel Release 77.9%2 blog.google · Deep Dive claim 04, verified
Model Release·Ars Technica·30 SEP 2026

Google Limits Gemini 4 Argon to Cyber Partners

On Sept 30, Google released Gemini 4 Argon, what the company calls its largest Gemini model, only to vetted cybersecurity partners through its Fairwind program, citing hacking risks. Google's comparison table shows Argon leading or tying on 14 of 19 evaluations and scoring 53 on the Artificial Analysis Intelligence Index, matching OpenAI's GPT-6 Astra.

Why it mattersCompare Argon's 53 index score with GPT-6 Astra's identical 53: Astra is buyable today, and Google has given no public date or price for Argon.

Continue reading
World

Donald Trump announced South Korea's US investment plan, which includes a $54 billion3 LNG development in Alaska and eight large nuclear power plants.

Markets

Sentiment reads Greed: the Nasdaq rose and the S&P 500 edged lower Wednesday after softer-than-expected inflation data, and both indexes closed a second straight quarterly gain.

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 30 SEP 2026 · READ 2026-10-01 · Yahoo Finance
S&P 500 7,651.547 -0.25%7 6D, window short 7,651.54 low / 7,743.41 high
Nasdaq 26,861.068 +0.24%8 6D, window short 26,797.54 low / 27,068.72 high
Bitcoin $83,410.329 -0.25%9 6D, window short $83,410.32 low / $84,458.09 high
Ethereum $2,679.6910 +0.11%10 6D, window short $2,676.72 low / $2,695.21 high
Crude Oil (WTI) $90.4311 +1.17%11 6D, window short $89.38 low / $94.61 high
Crypto Fear & Greed 74 / 10012 Greed 0 to 100 index 25 fear / 50 neutral / 75 greed · alternative.me
Today's Focus

What else the day turns on.

01

Meta Muse Targets Small Businesses

Meta announced Muse for Small Business on Sept 29, and the agent targets small businesses with goal-level delegation. Owners can hand Muse a broad goal, such as running the business or finding new customers, and the agent works to complete it.

02

OpenAI Disrupted Distillation Campaign

OpenAI disrupted a coordinated campaign that aimed to distill its models and published an account of the effort. The post outlines the security challenges involved and the countermeasures OpenAI used, which puts model distillation alongside other security threats the company defends against.

Deep Dive

Today's editorial.

All editorials
Deep Dive cover image: Google Shipped Its Top Model to Almost Nobody
Deep Dive7 min read

Google Shipped Its Top Model to Almost Nobody

Read full analysis
The Wire

6 stories from 6 desks.

Lead carried above
Benchmark·Venturebeat·30 SEP 2026

Argon's Coding Lead Depends on the Benchmark

Argon scored approximately 77.9%2 on DeepSWE v1.1 and approximately 57%4 on Terminal-Bench 4.0. It trails GPT-6 Astra by 10.5 points on FrontierSWE v2 and Claude Opus 5.5 by 9 points on Terminal-Bench 4.0.

Why it mattersThe leader flips depending on the coding suite, so the right model for an agent workload depends on which benchmark resembles it, which favors routing across vendors.

Read
Agents·CNBC·29 SEP 2026

Meta Brings Muse Agent to Small Businesses

Meta announced Muse for Small Business on Sept 29. Owners can give Muse a goal, such as running the business or finding new customers, and the agent works to complete it.

Why it mattersMarketing agencies serving small Meta advertisers should audit which client lead-generation work Muse could absorb before their next contract renewals.

Read
Model Release·Tech-ish·01 OCT 2026

Argon Writes Up to 1 Million5 Tokens

Gemini 4 Argon can reportedly generate up to 1 million5 tokens in a single output, up from a approximately 64,0006-token cap in the preceding Gemini generation.

Why it mattersA 1-million-token output is about 15.6 times the old cap; check Google's per-output-token price, once published, before planning long generation jobs.

Read
Also on the wire
  1. OpenAIOpenAI Disrupts Coordinated Model-Distillation Campaign
  2. UA.News (CNBC)Argon Runs Inside Google Before Wider Release
  3. ReutersTraders Doubt GPT-6.1 Astra Ships in October
The Globe

2 stories off the AI beat.

Infrastructure·Reuters·30 SEP 2026

South Korea Pledges Alaska LNG, Eight US Reactors

Donald Trump announced South Korea's US investment plan, including a $54 billion3 LNG development in Alaska and eight large nuclear power plants, alongside a gas plant in Texas that was reportedly agreed to previously.

Why it mattersEight large reactors draw on the same turbines, forgings and grid gear that data-center power projects need, so US equipment lead times are likely to stretch further.

Read
Also on the globe
  1. ReutersNasdaq Rises on Softer Inflation Data
The Lab

Tool of the day, field tested.

All Lab reports
Iris.ai home page, captured for the 1 October 2026 Lab report Tool of the day 4.613/ 10
AI Platform Pricing not published

Iris.ai

Iris.ai promises AI answers that hold up where a wrong one is unacceptable, and it keeps the product behind a demo request until you look like a buyer.

Iris.ai's pitch is control over what a model reads, backed by expert review loops that keep every answer traceable to a source, and none of it can be inspected without a sales call. Regulated R&D groups with a defined corpus and a procurement budget should pilot it; individual researchers will get further, faster, with a self-serve tool.

Capability 6.0
Ease 3.5
Value 4.0
Momentum 4.0

The only review on its Product Hunt page rates it 5 out of 5.Paraphrased · Product Hunt

Read the Lab report Try it
Also on the radar
Build

Basedash: build BI dashboards by describing them

Basedash is an AI-native business intelligence platform. You type questions in plain language, and it returns dashboards and AI-generated visualizations without any SQL. It lists as free, so connect one customer or revenue table first and ask the three questions your team asks every week. Before you share a chart, check the numbers against a query you already trust.

Try it
Productivity

AgentLaunch: an agent that submits your product to directories

AgentLaunch is an AI agent that submits startups, SaaS products and AI tools to online directories, working from a catalog of 397 entries according to its There's An AI For That listing. Write your product description, category and screenshots once before you start, so every submission stays consistent. Then track which listings actually send referral traffic and drop the rest from your next launch.

Try it
Productivity

Scisummary v1.4.0: summaries of scientific articles in minutes

SciSummary uses AI to condense scientific articles so you can decide quickly whether a paper deserves a full read. It holds a 3.7 rating from 38 reviews on There's An AI For That, so use it to triage papers and keep the original for anything you cite. Spot-check each summary's methods and sample size against the paper itself.

Try it
Productivity

Superhuman AI: AI features built to reach inbox zero faster

Superhuman AI is a set of AI features inside the Superhuman email client, aimed at cutting the time you spend triaging and replying. Try it on your highest-volume folder for one week and count how many drafts you send with light edits. That count tells you whether it is saving real time.

Try it
The Arena

3 labs moved this week.

  1. Google DeepMindAnnounces Argon, the Gemini 4 flagship, with early access limited to cybersecurity partners businesstimes.com.sg · 30 SEP 202601
  2. OpenAISays it disrupted a coordinated model-distillation campaign as ChatGPT suffers an outage openai.com02
  3. Meta AIRecent Joins voluntary frontier-safety pact while OpenAI's Dots targets its Muse assistant manilatimes.net · 01 OCT 202603
THE DOJO · BUILD TODAY

Test the models you can actually buy on your own work

  1. 01

    Run your own coding eval. Argon's lead swings by benchmark, so pull 20 real tickets from your repo and score GPT-6 Astra and Claude Opus 5.5 on them before trusting any vendor table.

  2. 02

    Pilot one goal-level agent task. Give an agent such as Dots or Muse a single bounded business goal, like drafting outreach to new customers, and log every app it touches and every step it takes.

  3. 03

    Audit your API for distillation patterns. OpenAI just described disrupting a coordinated distillation campaign, so review your own endpoints for high-volume, systematic query bursts from single accounts and set rate alerts.

The Bottom Line

Argon ties Astra at 53, and only Astra ships today.

Google says Gemini 4 Argon matches GPT-6 Astra on the Artificial Analysis Intelligence Index, yet Argon has no public date or price while OpenAI is already building Dots agents on top of Astra. That gap means the equal score changes nothing about what you can deploy this quarter. Build on the model you can buy, keep your evals portable, and rerun them when Google opens Argon beyond Fairwind partners.

Receipts

Every number, and where it came from.

22 of 23 figures cited, 13 receipts: 0 verified, 7 reported, 6 estimated
  1. 014,000...that run on their own cloud computer and connect to over 4,000 apps including ChatGPT, Slack and TeamsOpenAI, via TLDR30 SEP 2026as publishedE
  2. 0277.9%Argon scored approximately 77.9% on DeepSWE v1.1 and approximately 57% on Terminal-Bench 4.0.Venturebeat30 SEP 202603B search corroborationE
  3. 03$54 billion...announced South Korea's US investment plan, including a $54 billion LNG development in Alaska and eight large nuclear...Reuters30 SEP 2026as publishedR
  4. 0457%Argon scored approximately 77.9% on DeepSWE v1.1 and approximately 57% on Terminal-Bench 4.0.Venturebeat30 SEP 202603B search corroborationE
  5. 051 MillionArgon Writes Up to 1 Million Tokens.Tech-ish01 OCT 2026as publishedE
  6. 0664,000...million tokens in a single output, up from a approximately 64,000-token cap in the preceding Gemini generation.Tech-ish01 OCT 202603B search corroborationE
  7. 077,651.54S&P 500 close, -0.25% on the sessionYahoo Finance30 SEP 2026yfinance closeR
  8. 0826,861.06Nasdaq close, +0.24% on the sessionYahoo Finance30 SEP 2026yfinance closeR
  9. 09$83,410.32Bitcoin close, -0.25% on the sessionYahoo Finance01 OCT 2026yfinance closeR
  10. 10$2,679.69Ethereum close, +0.11% on the sessionYahoo Finance01 OCT 2026yfinance closeR
  11. 11$90.43Crude Oil (WTI) close, +1.17% on the sessionYahoo Finance30 SEP 2026yfinance closeR
  12. 1274 / 100Crypto Fear & Greed close, Greed on the sessionalternative.me01 OCT 2026as publishedR
  13. 134.6 / 10Iris.ai, Koda score across four dimensions04R dossier01 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

188 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.