K Koda Intelligence
Subscribe
Cover image for the 07 October 2026 Signal: Nicknamed Le Chonk, Mistral's trillion-parameter Large 4 activates just 49 billion parameters, with public weights due October 27.
THE SIGNAL · 07 OCT 2026 · 9 MIN READ

Mistral ships Le Chonk, a trillion-parameter Large 4 preview

Mistral opened a preview API for Mistral Large 4 on October 6: 1 trillion parameters with 49 billion1 active, and public weights due October 27.

KI Koda Editorial · Updated 01:35 GMT · 21 sources
AIOPEN WEIGHTSAGENTS
24 of 32 figures cited15 receipts in the register
  1. 49 billion1CNBC
  2. 4,0002CNBC
  3. 501B3CNBC
The register
Lead Story

The day's defining move.

Model Release · CNBC
Lead storyModel Release 49 billion1 CNBC · 06 OCT 2026
Model Release·CNBC·06 OCT 2026

Mistral Ships 1-Trillion-Parameter Large 4 Preview

Mistral AI opened a preview API for Mistral Large 4, nicknamed Le Chonk, on October 6: a 1-trillion-parameter multimodal model with 49 billion1 active parameters, trained for two months on 4,0002 Nvidia Grace Blackwell GPUs. Weights go public October 27. It lands the same week as Reflection's Beam-501B3-A23B and Z.ai's GLM 5.3 on Amazon Bedrock.

Why it mattersTeams weighing self-hosting can test Mistral Large 4 through the preview API now and hold hardware commitments until the weights go public on October 27.

Continue reading
World

Iran has ramped up attacks on ships in the Strait of Hormuz, with nine reported in October so far, just as oil and gas flows had been climbing back toward prewar levels.

Markets

The S&P 500 rose 0.58% and WTI crude gained 0.74%, while Bitcoin slipped 0.33% and Ether fell 0.51%; the split leaves the crypto Fear & Greed index at 71 (Greed) reading hotter than crypto prices did on the session.

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 · Infographic
Visual

Intelligence map

Markets

6 levels at the close.

Market TerminalLAST CLOSE 06 OCT 2026 · READ 2026-10-07 · Yahoo Finance
S&P 500 7,818.939 +0.58%9 6D, window short 7,651.54 low / 7,818.93 high
Nasdaq 27,599.8910 +0.45%10 6D, window short 26,797.54 low / 27,599.89 high
Bitcoin $85,505.0711 -0.33%11 6D, window short $84,497.21 low / $86,480.30 high
Ethereum $2,697.2612 -0.51%12 6D, window short $2,668.15 low / $2,726.51 high
Crude Oil (WTI) $90.0913 +0.74%13 6D, window short $89.38 low / $92.87 high
Crypto Fear & Greed 71 / 10014 Greed 0 to 100 index 25 fear / 50 neutral / 75 greed · alternative.me
Today's Focus

What else the day turns on.

01

Ghost AI Unveils Core

Ghost AI unveils Core, a personal computer offering 24/7 AI features through on-device models, after founder Zain Javaid, a 19-year-old former quantitative analyst, raised $11M. Core completes tasks across connected apps and devices and remembers key details about the user over time.

02

CFTC Proposes Crypto Trading Framework

The CFTC proposes a federal framework for crypto trading, opening rulemaking on Regulation CTX and Regulation CAM to create a pathway for leveraged and margined retail crypto trading. The proposal would also establish a new 'crypto asset market' category of CFTC-registered exchanges.

Deep Dive

Today's editorial.

All editorials
Deep Dive cover image: Le Chonk Uses 4.7% of Its Trillion Parameters
Deep Dive6 min read

Le Chonk Uses 4.7% of Its Trillion Parameters

Read full analysis
The Wire

6 stories from 5 desks.

Lead carried above
Policy·Anthropic·06 OCT 2026

Anthropic Widens Cyber Verification Program Access

Anthropic expanded its Cyber Verification Program, opening its most capable models to more security teams. Covered models include Claude Opus 5.5, Claude Sonnet 5.5 and Mythos 5.1, plus future releases.

Why it mattersSecurity teams shut out of Anthropic's top models should apply to the expanded Cyber Verification Program now, since approval also covers future releases beyond Mythos 5.1.

Read
Enterprise·TechCrunch·05 OCT 2026

OpenAI Puts Visual Ads Beside Image Results

OpenAI expanded ChatGPT ads beyond their existing formats: visual ads now appear alongside image generation results, backed by new measurement tools for advertisers.

Why it mattersImage generation is costly to serve, so selling ads beside it lets OpenAI offset free-tier compute; brand teams need placement rules for unpredictable images.

Read
Enterprise·Bloomberg·05 OCT 2026

OpenAI Talks $30 Billion4 Round With UAE Funds

OpenAI is in talks to raise $30 billion4 from UAE funds and BlackRock, per TLDR.

Why it mattersFresh capital at this scale mostly buys compute, so a closed round would extend OpenAI's runway for model training and data center contracts into 2027.

Read
Also on the wire
  1. CNBCLawsuit Challenges Nvidia's $20 Billion5 Groq Deal
  2. BloombergUtah Pilot Lets AI Write Acne Prescriptions
  3. OpenAIAtlassian and OpenAI Expand Partnership
The Globe

3 stories off the AI beat.

Conflict·Bloomberg·06 OCT 2026

Iran Steps Up Hormuz Ship Attacks

Oil and gas flows through the Strait of Hormuz had been climbing back toward prewar levels; Iran has now ramped up attacks on ships, with nine reported in October so far.

Why it mattersHormuz carries a major share of global oil and gas exports, so each attack feeds directly into energy price swings and the power contracts that run data centers.

Read
Also on the globe
  1. ReutersVance Demands Iranian Enrichment Cuts
  2. LatimesUkraine Fires Nearly 900 Drones at Russia
The Lab

Tool of the day, researched.

All Lab reports
Zyte MCP home page, captured for the 7 October 2026 Lab report Tool of the day 6.615/ 10
Coding

Zyte MCP

Zyte's docs show one add command and one login putting its scraping API and your Scrapy Cloud projects inside a coding agent, with every request billed to a key you can cap.

Zyte MCP hands an agent page fetching, structured extraction, web search and Scrapy Cloud control through one remote endpoint, billed to a dedicated mcp_access_key you can track and limit. The catch is the meter: Zyte's pricing page says request cost depends on the target website and selected features, so a crawl's price stays unknown until you run the estimator or a small test. Scrapy users and builders who want agent-driven extraction under a spending ceiling should adopt it now.

Capability 7.0
Ease 8.0
Value 6.0
Momentum 5.0
Read the Lab report Try it
Also on the radar
Build

Labelf AI: train text classifiers without writing code

Labelf AI is a low-code NLP platform that builds text classification models through active learning. The model asks you to label the examples it is least sure about, so you reach a usable classifier with fewer labels. Start with something like support tickets or survey responses on the free option, label a few batches, and check accuracy before you put it in front of a real workflow.

Try it
Build

Humalike: give your AI agents social skills and initiative

Humalike sells behavioral infrastructure, delivered as APIs and models, that makes AI agents act more like people. Its focus is social skills and proactiveness, such as knowing when to speak up instead of waiting for a prompt. It is enterprise-only, so teams building customer-facing agents should request a pilot and compare conversation quality against their current agent on a fixed set of real transcripts.

Try it
Productivity

Synter Media AI: agents that plan and optimize your ad campaigns

Synter Media AI is an agentic advertising platform that plans, launches, and optimizes campaigns across Google Ads, Meta, LinkedIn, Reddit, and X from one place. Connect one or two channels first and give the agent a capped test budget. Review its optimization changes against your manual baseline before you hand it more spend. Pricing is not published, so confirm the cost model before committing.

Try it
Productivity

Epismo: hand off AI work without re-briefing anyone

Epismo is a free collaboration workspace that bundles the context, decisions, and outputs for a piece of AI work into organized cases. You can switch between AI tools or teammates without repeating the brief. Its Auto Review feature gives a saved result a fresh review and flags issues while leaving the original untouched. Use it for multi-step projects where several agents or people touch the same deliverable.

Try it
The Arena

4 labs moved this week.

  1. MistralMistral Large 4, a 1-trillion-parameter open-weight model, enters developer and government preview cnbc.com · 06 OCT 202601
  2. OpenAIOpenAI expands its Atlassian partnership and details new ChatGPT ad formats openai.com02
  3. AnthropicAnthropic expands its Cyber Verification Program anthropic.com · 06 OCT 202603
  4. Meta AIRecent Meta's Muse personal agent lifts AMD as investor attention shifts to agents cnbc.com · 06 OCT 202604
Builder Radar

1 receipt from the repos.

Primary sources, pulled by index
Receipts
  1. Behind: Anthropic Widens Cyber Verification Program Access Docs14604842 real time cyber safeguards on claude opus and sonnet

    We'll soon be expanding the Cyber Verification Program to include Opus 5.5, Sonnet 5.5, and Mythos class models. Enable data retention on Claude on Google Cloud. In order to access through you must consent to and configure data retention.

    support.claude.com

Receipts come from the Firecrawl Developer Index (READMEs, docs, issues, merged PRs), first-party sources only. Quotes are verbatim.

THE DOJO · BUILD TODAY

Budget for memory and meters before you scale anything

  1. 01

    Size the memory bill for Large 4. Plan for roughly a terabyte at 8-bit and price that footprint across your hosting options before you treat the 49-billion active count as the cost driver.

  2. 02

    Give Zyte MCP its own key. Issue a dedicated mcp_access_key for your agent, set a usage limit on it, and review the tracked spend after the first real crawl.

  3. 03

    Check Anthropic's Cyber Verification Program. If you run a security team, find out whether the expanded program now covers you for access to Claude Opus 5.5 and Mythos 5.1.

The Bottom Line

Price the memory before the model.

Mistral's Le Chonk and Zyte's MCP endpoint point at the same lesson: the expensive part of an AI stack often sits in a line item that the headline number hides. For Large 4 that line item is the terabyte of memory needed to hold the model. For an agent that browses the web, it is the meter on every fetch. Put a budget on each one before you scale.

Receipts

Every number, and where it came from.

24 of 32 figures cited, 15 receipts: 2 verified, 12 reported, 1 estimated
  1. 0149 billion...October 6: a 1-trillion-parameter multimodal model with 49 billion active parameters, trained for two months on 4,000...CNBC06 OCT 2026as publishedR
  2. 024,000...49 billion active parameters, trained for two months on 4,000 Nvidia Grace Blackwell GPUs.CNBC06 OCT 2026as publishedR
  3. 03501BIt lands the same week as Reflection's Beam-501B-A23B and Z.ai's GLM 5.3 on Amazon Bedrock.CNBC06 OCT 2026as publishedR
  4. 04$30 BillionOpenAI Talks $30 Billion Round With UAE Funds.Bloomberg05 OCT 202603B search corroborationV
  5. 05$20 BillionLawsuit Challenges Nvidia's $20 Billion Groq Deal.CNBC05 OCT 202603B search corroborationV
  6. 06501BReflection unveils Beam, a 501B-parameter model it says rivals leading Chinese models on substantially less computeReflection, via TLDR06 OCT 2026as publishedR
  7. 07$28.7 millionStrategy buys 334 BTC for $28.7 million as total holdings top 848,000 BTCStrategy, via TLDR06 OCT 2026as publishedR
  8. 08848,000Strategy buys 334 BTC for $28.7 million as total holdings top 848,000 BTCStrategy, via TLDR06 OCT 2026as publishedR
  9. 097,818.93S&P 500 close, +0.58% on the sessionYahoo Finance06 OCT 2026yfinance closeR
  10. 1027,599.89Nasdaq close, +0.45% on the sessionYahoo Finance06 OCT 2026yfinance closeR
  11. 11$85,505.07Bitcoin close, -0.33% on the sessionYahoo Finance07 OCT 2026yfinance closeR
  12. 12$2,697.26Ethereum close, -0.51% on the sessionYahoo Finance07 OCT 2026yfinance closeR
  13. 13$90.09Crude Oil (WTI) close, +0.74% on the sessionYahoo Finance06 OCT 2026yfinance closeR
  14. 1471 / 100Crypto Fear & Greed close, Greed on the sessionalternative.me07 OCT 2026as publishedR
  15. 156.6 / 10Zyte MCP, Koda score across four dimensions04R dossier07 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

194 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.