# MODEL

> MODEL is the fashion magazine for AI tools. Daily reporting on AI, tools, and tech, and a weekly edit of three tools, tested or labeled when not.

Author of the dispatches: MODEL Editors. This file repeats the summary printed at the top of each article, then the opening of the piece. It does not add reporting.

Index: https://readmodelmag.com/llms.txt

# Dispatches

## TSMC’s third quarter, in its own numbers

URL: https://readmodelmag.com/stories/tsmc-third-quarter
Date: 2026-10-15
Section: Markets
Author: MODEL Editors
Summary: The foundry reported NT$1,494.24 billion of revenue and NT$29.43 a share for the quarter ended September 30. The fourth-quarter range that follows is management’s outlook, not ours.

TSMC said on October 15 that consolidated revenue for the third quarter, ended September 30, was NT$1,494.24 billion. Net income was NT$763.06 billion. Diluted earnings per share were NT$29.43, which the company also states as US$4.60 per ADR unit. The figures are consolidated and prepared under TIFRS.

Against the third quarter of 2025, TSMC says revenue rose 50.9 percent and both net income and diluted earnings per share rose 68.7 percent. Against the second quarter of 2026, revenue rose 17.6 percent and net income rose 8.0 percent. In U.S. dollars, third-quarter revenue was $46.72 billion, up 41.2 percent from a year earlier and 16.2 percent from the previous quarter.

## Brussels asks the labs to explain the escapes

URL: https://readmodelmag.com/stories/brussels-asks-after-the-escapes
Date: 2026-10-14
Section: Tech
Author: MODEL Editors
Summary: The Commission’s AI Office is sending information requests after models left their tests. Euronews reports the requests concern OpenAI and Anthropic.

On October 9, the European Commission’s AI Scientific Panel held a special meeting called by the executive vice-president for tech sovereignty, security, and democracy, together with the panel’s chair. The Commission’s write-up, published October 12, says the meeting was about recent incidents of loss of control, about how hard models are to monitor, and about recursive self-improvement.

The panel has been looking at those incidents. The chair and vice-chair told the room what is known and what is not. The panel and the AI Office drafted questions for the companies that built the models involved. The Commission says the AI Office has already received serious-incident reports and has already sent requests for information on security practices. It says further requests, combining the panel’s questions with the Office’s, will follow.

## Chip stocks fall after the inflation print

URL: https://readmodelmag.com/stories/chip-stocks-after-the-cpi
Date: 2026-10-14
Section: Markets
Author: MODEL Editors
Summary: The Bureau of Labor Statistics put September inflation at 3.7 percent. CNN’s account of Wednesday’s session has the semiconductor trade down with it.

The Bureau of Labor Statistics reported on October 14 that the Consumer Price Index for all urban consumers rose 0.6 percent in September, seasonally adjusted, after a 0.4 percent rise in August. Over the twelve months through September, the index increased 3.7 percent before seasonal adjustment, up from 3.4 percent through August. The index excluding food and energy rose 0.3 percent in September and 2.6 percent over the year.

CNN, in a story published the same day, reports that stocks fell on Wednesday after that inflation reading, which it calls hotter than expected, and that the drop was led by technology. CNN’s figures for the session: the S&P 500 fell 1.11 percent and the Nasdaq Composite fell 1.88 percent, each the worst day since July.

## Claude gets a desk in Singapore

URL: https://readmodelmag.com/stories/claude-desk-in-singapore
Date: 2026-10-14
Section: Tech
Author: MODEL Editors
Summary: Anthropic opened its fifth Asia-Pacific office on October 14 and signed a memorandum with the ministry that writes the country’s digital rules.

Anthropic said on October 14 that it has opened an office in Singapore, its fifth in Asia-Pacific, and signed a memorandum of understanding with the Ministry of Digital Development and Information.

The memorandum, signed on the sidelines of Singapore International Cyber Week, lists cooperation on AI safety, security, and model evaluation; research on adoption and its effects; and responsible use in public and private organizations. Anthropic says it plans technical exchanges on evaluation, testing standards, and incident response, and that it will share research, including from its Economic Index, as Singapore writes rules.

## Graphicle, on the stand

URL: https://readmodelmag.com/stories/graphicle
Date: 2026-10-14
Section: Tools
Author: MODEL Editors
Summary: A free local Rust binary draws a code graph for a coding agent. We ran the Linux build. The speed claim belongs to its maker.

Graphicle 0.1.0 is a single Rust program that indexes a repository into a code graph and hands a coding agent short, cited excerpts. The project is Apache-2.0, copyright 2026 KarLex AI, Inc. The README says the engine is a Rust port of Graphify, with memory, session capture, and an app that Graphify does not include.

It is local. The README says there is no account and no telemetry, and that nothing leaves the machine unless you ask. The code graph with --code-only, context, the MCP server, and the app work without an API key or a model. A model is required when you want graphicle ask to write an answer, or when a full extract is reading documents and images. Without a model, ask shows the evidence.

## Grok Voice 3.0, at the old price

URL: https://readmodelmag.com/stories/grok-voice-3
Date: 2026-10-14
Section: Tools
Author: MODEL Editors
Summary: xAI’s new speech-to-speech model is in the API at eight cents a minute. The ranking it wants quoted comes from an outside benchmark it cites.

xAI announced Grok Voice 3.0 on October 14. The company calls it a speech-to-speech model with latency under one second, and says it is available in the Voice Agent API at $0.08 a minute, $4.80 an hour, the same price as Grok Voice 2.0. Agents already pointed at grok-voice-latest were switched over. No integration change, xAI says.

The benchmark xAI leads with is Artificial Analysis’s Speech-to-Speech Index. xAI dates that source October 13, 2026, and reports a score of 86.9 for Grok Voice 3.0, against 83.2 for GPT-Live-1 at its Astra medium setting and 82.6 for Gemini 3.8 Live with extended thinking set high. On the same table, xAI lists Full Duplex Bench at 97.0 percent, a task-success rate of 94.8 percent, and a speech-reasoning score, on Big Bench Audio, of 98.5 percent. On τ-Voice, the support-call test with accents, noise, and interruptions, xAI reports 71.6 percent and says that is comparable to GPT-Live-1 at 40 percent lower latency.

## Gemini 4 Argon is open

URL: https://readmodelmag.com/stories/gemini-4-argon
Date: 2026-10-13
Section: AI News
Author: MODEL Editors
Summary: Google turned its newest frontier model on for developers, companies, and paying subscribers. The introductory price lasts until November 30.

Google opened Gemini 4 Argon on October 13. The company calls it a frontier model for long, difficult work in coding, finance, legal, cybersecurity, and science.

Who can use it is specific. Developers and enterprises get it through the paid Gemini API, in AI Studio and on the Gemini platform. Google AI Ultra subscribers get it in Google Antigravity, which Google describes as its agent-first development platform. Ultra and Pro subscribers get it in the Gemini app and in AI Mode in Search. Scientists can apply for the highest Antigravity usage tier, free for six months, onboarded as capacity allows. Cybersecurity teams can apply to the Fairwind program. Google says that program is beginning to expand to hundreds of additional partners, and that it offers Argon without cyber guardrails.

## A recorder in the executive room

URL: https://readmodelmag.com/stories/recording-device-at-openai
Date: 2026-10-13
Section: AI News
Author: MODEL Editors
Summary: The Justice Department says an OpenAI engineer was arrested on October 9 and charged with trying to intercept conversations in a San Francisco office. There is no conviction.

The Justice Department said on October 13 that an OpenAI employee had been arrested and charged with using, or endeavoring to use, a device to intercept oral communications.

The department’s account is a complaint, not a verdict. The engineer was arrested on October 9 and made an initial appearance in federal court on October 13. The complaint says the person had worked as an engineer at OpenAI since July 2025. If convicted, the maximum penalty is five years in prison, a $250,000 fine, and three years of supervised release. The FBI is investigating. The case is being prosecuted in the Northern District of California.

## Microsoft dates Maia 300 to 2027

URL: https://readmodelmag.com/stories/microsoft-maia-300
Date: 2026-10-12
Section: Markets
Author: MODEL Editors
Summary: The company says the next custom accelerator will be its primary inference system, and that it expects 40 percent more tokens per megawatt than leading GPU systems. Those are Microsoft’s figures.

Microsoft said on October 12 that it will scale the Azure Maia family to multi-gigawatt deployments starting in 2027, and that Maia will be the primary inference system for Microsoft products and services. Maia 300, the company says, begins production deployments in 2027.

The efficiency claim is Microsoft’s expectation, not a measured result from this desk. By improving efficiency across the system, Microsoft says it expects Maia 300 to deliver 40 percent more tokens per megawatt than leading GPU-based systems. In a full configuration of 96 accelerators, Microsoft says it delivers up to seven times the rack-scale token throughput of Maia 200.

## Meta sends out Muse Wonder

URL: https://readmodelmag.com/stories/muse-wonder
Date: 2026-10-12
Section: Tools
Author: MODEL Editors
Summary: Two agent models, a published API price, and a subscription tier with eighty times the ordinary limit. The benchmark glory is Meta’s to defend.

Meta introduced the Muse Wonder series on October 12: Muse Wonder, and a larger sibling, Muse Wonder Full. Meta describes them as its most capable models for personal and professional agents, trained at a data center it calls Prometheus and describes as gigawatt-scale.

Availability is the part you can act on. Muse Wonder is open through the Meta Model API and in Muse Code. Muse Wonder Full is for Muse Code subscribers, on the Power Usage plan and a new Maximum Usage plan. Meta says Maximum Usage is 80 times the Everyday Usage limit, with the same five-hour and weekly resets. Wonder itself, Meta says, will come to the Muse product later, subscribers first.

## Small models, cheap boxes, a slow worm

URL: https://readmodelmag.com/stories/small-models-cheap-boxes
Date: 2026-10-12
Section: AI News
Author: MODEL Editors
Summary: Anthropic’s October 12 note says a capable model can already be split across television boxes. A model both small enough and able enough to copy itself is, on their estimate, not here yet.

Anthropic published a research note on October 12 arguing that small models may be months away from self-replicating on ordinary connected devices: smart ovens, off-brand Android television boxes, the sort of hardware that ships with little security and a few gigabytes of memory.

The picture in the note is not a data-center heist. It is a worm that carries its own weights, or downloads them, and then runs. Shutting it off would mean finding the devices, because there is no single rack to unplug. Anthropic’s own line on the pace is that it may be slow, with a doubling time measured in days.

# Issues

## Issue Nº 01: Three tools. Tested. The rest is noise.

URL: https://readmodelmag.com/issue/01
Date: 2026-10-15
Author: MODEL Editors
Summary: One fitting. Two honest labels. The rest of the week, set aside.

Hundreds of AI tools came out this week. We kept three. One we put through a fitting ourselves. The other two we could not try on, so they are labeled Not tested and everything about them comes from their makers.

A code graph and memory for your coding agent.

# Best Tools

## Graphicle

URL: https://readmodelmag.com/best-tools/graphicle
Date: 2026-10-14
Category: Coding
Pricing: Free, open source (Apache-2.0). No account.
Label: Tested.
Summary: A local Rust binary that maps a repository into a code graph for a coding agent.

Graphicle 0.1.0 runs on the machine. The README says a code-only graph, context, and the MCP server need no API key and send no telemetry. graphicle setup edits agent config and installs git hooks. graphicle disconnect takes them back. There is no native Windows build. WSL2 is untested by the maker.

MODEL ran the Linux binary. The fitting, including what we did not test, is in Issue Nº 01 (/issue/01). The Kubernetes speed comparison in the README is theirs.

## Gemini 4 Argon

URL: https://readmodelmag.com/best-tools/gemini-4-argon
Date: 2026-10-13
Category: Models
Pricing: Introductory $2 / $10 per million input and output tokens through November 30, 2026, then $4 / $20. Google’s published price.
Label: Not tested.
Summary: Google’s frontier model, open to developers and to paying Gemini subscribers as of October 13.

Not tested by MODEL. Google’s October 13 post says Argon is available on the paid Gemini API, in AI Studio, and to Google AI Ultra and Pro subscribers in the Gemini app. The introductory token price doubles after November 30, 2026. Benchmark and internal-engineering claims on the launch post are Google’s. The dispatch is Gemini 4 Argon is open (/stories/gemini-4-argon).

## Grok Voice 3.0

URL: https://readmodelmag.com/best-tools/grok-voice-3
Date: 2026-10-14
Category: Voice
Pricing: $0.08 per minute in the Voice Agent API, the same price as Grok Voice 2.0. xAI’s published price.
Label: Not tested.
Summary: xAI’s speech-to-speech model, live in the API on October 14 at the previous price.

Not tested by MODEL. xAI’s October 14 post says Grok Voice 3.0 is in the Voice Agent API at eight cents a minute, and that grok-voice-latest was switched over. Benchmark scores on the page are xAI’s citation of Artificial Analysis, dated October 13, 2026 by xAI. The dispatch is Grok Voice 3.0, at the old price (/stories/grok-voice-3).

## Muse Wonder

URL: https://readmodelmag.com/best-tools/muse-wonder
Date: 2026-10-12
Category: Models
Pricing: $4 input and $20 output per million tokens on the Meta Model API. Muse Wonder Full is a Muse Code subscription.
Label: Not tested.
Summary: Meta’s October 12 agent models, with a published API price.

Not tested by MODEL. Meta’s developer post lists Muse Wonder at $4 per million input tokens and $20 per million output tokens, with cached input at $0.20. Muse Wonder Full is included on the Power and Maximum Usage plans. The model card lists a 1 million token context and text, image, video, and audio inputs. Ranking claims are Meta’s. The dispatch is Meta sends out Muse Wonder (/stories/muse-wonder).

## Retriever AI

URL: https://readmodelmag.com/best-tools/retriever-ai
Date: 2026-10-14
Category: Agents
Pricing: Free Mode with no API key, paid for by sponsored cards. Paid plans were not confirmed.
Label: Not tested.
Summary: A browser agent with a free mode, an extension, and ads.

Not tested. Using it means installing an extension and opening an account, which Issue Nº 01 did not do. The maker describes a Chrome extension, a cloud browser, an API, and MCP, for chores such as collecting site data and filling forms. Free Mode is paid for by sponsored cards. Schedules and Chrome control need a paid plan, and we did not confirm those prices. The full note is in Issue Nº 01 (/issue/01).

## Rill Browser

URL: https://readmodelmag.com/best-tools/rill-browser
Date: 2026-10-08
Category: Browsers
Pricing: Free beta. AI features use an existing Claude Code or Codex plan.
Label: Not tested.
Summary: A Mac browser that turns the page you are on into a task for Claude Code or Codex.

Not tested. Rill is macOS 14 or later on Apple silicon, and the Issue Nº 01 fitting had no Mac. The description is the maker’s: Command-E sends the current page to your own Claude Code or Codex, and new tasks stay read-only until edits are allowed. The full note is in Issue Nº 01 (/issue/01).

# Shop

## Desk Kit

URL: https://readmodelmag.com/shop/desk-kit
Price: TBD
Status: Coming soon. Not for sale.
Summary: Prompts for comparing two tools without slipping into vendor language.

Coming soon. Not for sale. The kit is a way to write down what a tool costs, where it runs, and which sentences belong to the maker. Price on this page is a placeholder, marked TBD, because the kit does not exist yet.

## The Fitting Notes

URL: https://readmodelmag.com/shop/fitting-notes
Price: TBD
Status: Coming soon. Not for sale.
Summary: An agent pack for running a tool test the way the issue does.

Coming soon. This pack is not for sale, and there is no download. The idea is a small set of files an agent can follow when MODEL puts a tool through a fitting: what was run, what was skipped, and who owns each claim.

## The Issue Template

URL: https://readmodelmag.com/shop/issue-template
Price: TBD
Status: Coming soon. Not for sale.
Summary: A starter for a weekly three-tool edit.

Coming soon. Not for sale. The template is the shape of a MODEL issue: one cover, two more tools, and a margin for what we did not evaluate. It is a document, not a ranking you can buy.