K Koda Intelligence
Subscribe
Cover image for the 13 October 2026 Signal: Meta is back at the frontier: Muse Wonder claims scores comparable to Claude Fable 5.1 and GPT-6 Astra.
THE SIGNAL · 13 OCT 2026 · 10 MIN READ

Meta's Muse Wonder claims frontier parity with Claude Fable 5.1 and GPT-6 Astra

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on October 12, putting Meta back in the frontier conversation. Google is already testing its next Gemini, so the comparison table will shift again soon.

KI Koda Editorial · Updated 01:34 GMT · 28 sources
AIMODELSBUILDERS
28 of 36 figures cited16 receipts in the register
  1. $41Wired
  2. $202Wired
  3. 3.38%3Wired
The register
Lead Story

The day's defining move.

Model Release · Wired
Lead storyModel Release $41 Wired · 12 OCT 2026
Model Release·Wired·12 OCT 2026

Meta Returns to the Frontier With Muse Wonder

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on October 12, claiming scores comparable to Anthropic's Claude Fable 5.1 and OpenAI's GPT-6 Astra. Muse Wonder costs $41 per million input tokens and $202 per million output tokens via the Meta Model API; Wonder Full is limited to Muse Code subscribers. Meta shares closed up 3.38%3 at $742.964.

Why it mattersPrice the $202-per-million output rate against Claude Fable 5.1 bills, knowing Wonder Full, the variant claiming the coding and legal wins, stays off the API.

Continue reading
Markets

The Nasdaq fell 0.44% and the S&P 500 slipped 0.06% as WTI crude gained 1.44%, while Ether dropped 4.60% and Bitcoin 2.86%; the crypto Fear & Greed index still reads 57 (Greed), so the gauge has not yet caught up with the session's crypto selloff.

Wild Card

Nvidia is working both ends of the AI stack at once: co-proposing the USFP optical-module standard with Cisco for AI data centers while weighing an outright purchase of Reflection AI, a company it already backs.

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 12 OCT 2026 · READ 2026-10-13 · Yahoo Finance
S&P 500 7,807.2410 -0.06%10 6D, window short 7,765.36 low / 7,818.93 high
Nasdaq 27,245.9011 -0.44%11 6D, window short 27,193.34 low / 27,599.79 high
Bitcoin $81,290.3112 -2.86%12 6D, window short $81,290.31 low / $83,679.47 high
Ethereum $2,418.2313 -4.60%13 6D, window short $2,418.23 low / $2,573.53 high
Crude Oil (WTI) $93.1714 +1.44%14 6D, window short $88.28 low / $93.17 high
Crypto Fear & Greed 57 / 10015 Greed 0 to 100 index 25 fear / 50 neutral / 75 greed · alternative.me
Today's Focus

What else the day turns on.

01

Atomic Machines' Matter Compiler

Jeff Holden, Amazon's 10th engineer and a builder of Amazon Prime and Uber's self-driving division, has emerged from six years in stealth with Atomic Machines, which has raised $250M. Its Matter Compiler is a general-purpose manufacturing system that turns code into working micro-machines, pitched as AI-native 'vibe manufacturing.'

02

HPE's Two-Year Ticket Forecast

HPE's networking boss says AI will handle all trouble tickets without humans in two years. The claim is specific and dated, which makes it a clear benchmark to measure against as that two-year window closes.

03

Seniors Lead Developer Hiring Rebound

Developer hiring is rebounding, but the rebound favors seniors while workers aged 22 to 25 stay under pressure. The split means a recovering market is still leaving its youngest entrants behind.

Deep Dive

Today's editorial.

All editorials
Deep Dive cover image: Nadella Says Assume Your AI Is Compromised
Deep Dive7 min read

Nadella Says Assume Your AI Is Compromised

Read full analysis
The Wire

7 stories from 6 desks.

Lead carried above
Hardware·NVIDIA

Nvidia and Cisco Propose USFP Optical Module Standard

400 Gb/s per electrical lane: that is the first product under USFP, a pluggable optical-module standard Nvidia and Cisco introduced for AI data centers, claiming four times existing bandwidth density.

Why it mattersData center network planners should ask optics vendors for USFP roadmaps before locking their next switch refresh, since denser pluggables change rack port counts.

Read
Trend·Theregister·09 OCT 2026

Global PC Shipments Fall 20%5 as Prices Rise

Global PC shipments dropped 20%5 as rising prices hammered demand.

Why it mattersSet the 20%5 decline against your fleet refresh quote: rising unit prices mean an AI PC rollout costs more per seat than last year's budget assumed.

Read
Enterprise·Bloomberg·09 OCT 2026

Startup Behind Jev Reaches $7.5 Billion6 Valuation

$7.5 billion6 is the new valuation for the startup behind Jev, as decision models gain popularity; Microsoft introduced Microsoft-Decision-1 for fast decision-making this week.

Why it mattersTeams evaluating Jev should test it against Microsoft-Decision-1 on speed and cost per call, since Microsoft can bundle its model into existing enterprise contracts.

Read
Also on the wire
  1. BloombergNvidia Weighs Buying Reflection AI Outright
  2. TechCrunchCloudflare Acquires Deno for Its Workers Platform
  3. The VergeMicrosoft CEO: Assume Your AI Is Compromised
  4. Tech XploreUCLA's PhoneBot Turns Old Smartphones Into Humanoids
The Globe

1 story off the AI beat.

Infrastructure·Dw·12 OCT 2026

Riyadh Airport Closure Extended Through October 17

Riyadh's King Khalid airport stays shut through October 17 after Houthi attacks that have killed at least 18.

Why it mattersTravelers and air-freight shippers should reroute bookings away from Riyadh's King Khalid airport until authorities confirm a firm reopening date.

Read
The Lab

Tool of the day, researched.

All Lab reports
Talorys home page, captured for the 13 October 2026 Lab report Tool of the day 6.716/ 10
Productivity Free

Talorys

Talorys packs chat, memory, tasks, notes and reminders into a Worker and a SQLite Durable Object on your own Cloudflare account, and the first thing worth testing is whether it actually remembers.

Talorys keeps every piece of a personal assistant, from chat to reminders, inside a Cloudflare account you control, and the price is the effort of running it yourself. Memory is the open question, with early users reporting ignored preferences, so trust it with routines before you trust it with anything important. Solo builders comfortable with Workers will get the most out of it; people who want someone else handling updates should look elsewhere.

Capability 6.0
Ease 6.0
Value 8.5
Momentum 6.5

Interest in the launch came from people who want an assistant without a monthly subscription or a vendor-run backend.Paraphrased · Hacker News

Read the Lab report Try it
Also on the radar
Productivity

Kai: an autonomous assistant for your inbox, calendar, and meetings

Kai is an AI executive assistant for solopreneurs, freelancers, and startup founders. It works across email, calendar, and tasks to triage your inbox, transcribe and summarize meetings, and manage follow-ups. Start the free trial (Product Hunt sign-ups get double trial credits), connect one client's email thread and calendar first, and check its triage decisions for a week before you hand over your whole inbox.

Try it
Creativity

adam.new: automate CAD edits and engineering tasks with AI

adam.new targets mechanical engineers and industrial designers who need quick model edits and automated design chores. It integrates with multiple CAD platforms and adds real-time analytics on your work. The freemium tier is enough to test it on one repetitive edit, such as resizing a part family. Paid options start at $209/month if it saves you time.

Try it
Build

Stage5.ai: describe a conversational assistant and deploy it without code

Stage5.ai is a no-code platform that designs, provisions, and deploys production-ready conversational assistants from a plain description of what you need. Write a tight spec that covers the assistant's audience, the questions it must answer, and what it should refuse. Then run edge-case prompts against the result before you put it in front of customers. Pricing is not listed, so confirm costs before you commit.

Try it
Hardware

Simplicity AI SDK: let your coding assistant build, flash, and debug IoT firmware

Silicon Labs' public-beta SDK gives AI coding assistants such as Copilot, Cursor, and Codex structured access to Silicon Labs SDKs, documentation, hardware, and build, flash, and debug workflows. Bluetooth LE and IoT developers can ask the assistant to scaffold a project and push it to a dev board without leaving the editor. Because it is in beta, keep a manual build path ready and check generated peripheral configuration against the datasheet.

Try it
The Arena

2 labs moved this week.

  1. Meta AIMeta launched Muse Wonder and Muse Wonder-Full on October 12, calling them its most capable AI models to date. bloomberg.com · 12 OCT 202601
  2. Google DeepMindGoogle DeepMind is a participant in a $1.8 billion Biohub initiative focused on open biological data and AI-enabled cell research. ground.news · 12 OCT 202602
Builder Radar

1 receipt from the repos, one paper.

Primary sources, pulled by index
Receipts
  1. Behind: Nvidia and Cisco Propose USFP Optical Module Standard DocsDeveloper Resources for Media and Entertainment

    NVIDIA Optical Flow SDK. Optical Flow SDK exposes the latest hardware capability of Turing and Ampere GPUs dedicated to computing the relative motion of pixels between images.

    developer.nvidia.com
Paper of the day

The Labeling Problem in Hallucination Detection Benchmarks: An Empirical Evaluation

Jorma Valjakka, Juhani Kivimäki, Juha Mylläri et al. · arXiv 2026-10-06

In recent years, several methods for detecting when large language models (LLMs) hallucinate have been developed. These methods are often benchmarked with open-domain question answering (QA) datasets containing questions and corresponding short reference answers.

Across judge families, p2 is usually the best or near-best prompt, indicating that the main empirical change occurs between p1 and p2 after which the effect of correctly specifying the target criterion saturates.
Read the paper

Receipts come from the Firecrawl Developer Index (READMEs, docs, issues, merged PRs), first-party sources only. The paper comes from the Firecrawl Research Index: arXiv, indexed in the last 7 days and submitted this month or last, picked by today's research theme. Quotes are verbatim.

THE DOJO · BUILD TODAY

Turn today's claims into numbers you can check.

  1. 01

    Date HPE's forecast. Write down the claim that AI will handle all trouble tickets without humans within two years, then track your own auto-resolved ticket share every month. Compare it to the line you drew today.

  2. 02

    Test a self-hosted assistant. Spin up Talorys in a Cloudflare account you control and give it a few facts on day one. Ask for them again a week later, since memory is where early users see the most risk.

  3. 03

    Audit your junior pipeline. Developer hiring is rebounding for seniors while workers aged 22 to 25 stay under pressure. Check whether your team has a mentoring path for early-career engineers, and add one if it does not.

The Bottom Line

Benchmarks move faster than your stack should.

Meta says Muse Wonder matches the leaders, and Google is already testing a Gemini 4 Argon checkpoint, so the ranking you pick today may not hold next month. Build a small eval set from your own tasks and run each new model against it before you change providers. Keep the model call behind one interface so a swap takes an afternoon. Only then let a vendor's scores influence what you ship.

Receipts

Every number, and where it came from.

28 of 36 figures cited, 16 receipts: 2 verified, 9 reported, 5 estimated
  1. 01$4Muse Wonder costs $4 per million input tokens and $20 per million output tokens via the Meta Model API; Wonder Full...Wired12 OCT 202603B search corroborationE
  2. 02$20Muse Wonder costs $4 per million input tokens and $20 per million output tokens via the Meta Model API; Wonder Full...Wired12 OCT 202603B search corroborationE
  3. 033.38%Meta shares closed up 3.38% at $742.96.Wired12 OCT 202603B search corroborationE
  4. 04$742.96Meta shares closed up 3.38% at $742.96.Wired12 OCT 202603B search corroborationE
  5. 0520%Global PC Shipments Fall 20% as Prices Rise.Theregister09 OCT 202603B search corroborationV
  6. 06$7.5 BillionStartup Behind Jev Reaches $7.5 Billion Valuation.Bloomberg09 OCT 202603B search corroborationV
  7. 07$7.5BTypeSafe AI hits a $7.5B valuation with $870M in funding one month after launching its Jev decision modelTypeSafe AI, via Superhuman12 OCT 2026as publishedR
  8. 08$870MTypeSafe AI hits a $7.5B valuation with $870M in funding one month after launching its Jev decision modelTypeSafe AI, via Superhuman12 OCT 2026as publishedR
  9. 09$20Paid options start at $20/month if it saves you time.adam.new: automate CAD edits and engineering tasks with AI13 OCT 2026as publishedR
  10. 107,807.24S&P 500 close, -0.06% on the sessionYahoo Finance12 OCT 2026yfinance closeR
  11. 1127,245.90Nasdaq close, -0.44% on the sessionYahoo Finance12 OCT 2026yfinance closeR
  12. 12$81,290.31Bitcoin close, -2.86% on the sessionYahoo Finance13 OCT 2026yfinance closeR
  13. 13$2,418.23Ethereum close, -4.60% on the sessionYahoo Finance13 OCT 2026yfinance closeR
  14. 14$93.17Crude Oil (WTI) close, +1.44% on the sessionYahoo Finance12 OCT 2026yfinance closeR
  15. 1557 / 100Crypto Fear & Greed close, Greed on the sessionalternative.me13 OCT 2026as publishedR
  16. 166.7 / 10Talorys, Koda score across four dimensions04R dossier13 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

200 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.