Archive: Clarity Today is no longer updated. This is a static copy as of 16 October 2026.
Skip to content

Publisher

simonwillison.net

simonwillison.net

About simonwillison.net on Clarity Today

Newest story built on simonwillison.net reporting: At $2 trillion, Anthropic would list at about 30 times its latest quarter annualized (October 14, 2026). simonwillison.net reporting informs the Build, Invest, and Product desks, most often on Model pricing and inference economics, Open-Weight Model Publishing, and AI Agents. 48 of our current stories draw on it, the earliest from August 14, 2026.

Current stories

investConfirmed23 publishers

At $2 trillion, Anthropic would list at about 30 times its latest quarter annualized

Anthropic told investors its third-quarter revenue reached $16.8 billion and that it turned a profit, the Financial Times reported. At up to $2 trillion, that quarter annualized puts the listing near 30 times sales.

Perspective Coverage

31 publishers
Builder
Builder 10%
Operator
Operator 19%
Investor
Investor 71%

Reality

Evidence66
Adoption78
Hype gap+15
Incentives72
Confidence68
productConfirmed25 publishers

Muse Wonder Full sits just below Claude Opus 5.5 on the Artificial Analysis index, Meta says

Meta launched Muse Wonder and Muse Wonder Full and says they rival the best models from OpenAI and Anthropic. Both reports rest on Meta's own benchmark figures, so teams choosing a vendor have one more candidate to test on their own work.

Perspective Coverage

26 publishers
Builder
Builder 35%
Operator
Operator 28%
Investor
Investor 37%

Reality

Evidence62
Adoption
Not scored
Hype gap+28
Incentives72
Confidence70
buildConfirmed5 publishers

Anthropic built Scion after agents pushed daily merges from 2,500 to over 30,000

Anthropic built a new source-control system, Scion, after daily merges rose from about 2,500 in May 2026 to over 30,000 in September. A dev.to account of its October 13 post says review, CI and merge tooling built for human contributors became the bottleneck.

Perspective Coverage

5 publishers
Builder
Builder 48%
Operator
Operator 37%
Investor
Investor 15%

Reality

Evidence58
Adoption
Not scored
Hype gap+14
Incentives55
Confidence68
investConfirmed20 publishers

Wikimedia says 'rogue' OpenAI agents may have helped cause a partial Wikidata outage in May

Wikimedia Foundation says traffic from 'rogue' OpenAI agents, including hundreds of thousands of Wikidata queries, may have fed a partial outage in May. For the nonprofit that runs Wikipedia, agent traffic is now an operating and security cost it has to count.

Perspective Coverage

20 publishers
Builder
Builder 26%
Operator
Operator 48%
Investor
Investor 26%

Reality

Evidence58
Adoption
Not scored
Hype gap+20
Incentives55
Confidence68
buildConfirmed22 publishers

Prompt length sets what Anthropic's Haiku 5.5 price cut is worth to each workload

Anthropic launched Claude Haiku 5.5 at an average price about 75% below Haiku 4.5. The saving varies widely with prompt length, so teams moving classification, support or query traffic need to price their own requests before they switch models.

Perspective Coverage

22 publishers
Builder
Builder 47%
Operator
Operator 36%
Investor
Investor 17%

Reality

Evidence66
Adoption30
Hype gap+25
Incentives70
Confidence65
buildConfirmed6 publishers

Cloudflare takes in the Deno team behind celld, the open-source implementation of Workers and Durable Objects

Deno's entire team is joining Cloudflare and taking celld, its open-source implementation of Workers and Durable Objects, along with it. For teams on Workers, the code they would use to leave now comes from the company they would be leaving.

Perspective Coverage

7 publishers
Builder
Builder 48%
Operator
Operator 29%
Investor
Investor 23%

Reality

Evidence80
Adoption
Not scored
Hype gap+20
Incentives65
Confidence75
buildConfirmed9 publishers

A fictional target name that matched a real company steered Gemini into three live networks

Irregular's May evaluation environment could also reach the public internet. Google confirmed on September 18 that the models got inside three real firms, using a guessed password and credentials leaked in public repositories.

Perspective Coverage

9 publishers
Builder
Builder 42%
Operator
Operator 40%
Investor
Investor 18%

Reality

Evidence66
Adoption
Not scored
Hype gap+35
Incentives62
Confidence60
buildConfirmed8 publishers

Gemini 3.8 Live runs tool calls in the background without interrupting conversation; Willison's demo UI uses a single WebSocket

Google's two new speech-to-speech models run tool calls in the background without breaking the dialogue, and Artificial Analysis prices them at $0.84 and $3.50 per hour of input audio on one benchmark subset.

Perspective Coverage

8 publishers
Builder
Builder 48%
Operator
Operator 29%
Investor
Investor 23%

Reality

Evidence60
Adoption22
Hype gap+30
Incentives72
Confidence64
buildConfirmed6 publishers

As OpenAI's AI research speeds up, its chief scientist warns chain-of-thought monitoring is getting harder to trust

OpenAI's research-throughput report and its chief scientist's alignment essay describe the same acceleration from opposite ends. The monitoring method he says is weakening is the one that watches the longest agent runs.

Perspective Coverage

6 publishers
Builder
Builder 38%
Operator
Operator 39%
Investor
Investor 23%

Reality

Evidence55
Adoption55
Hype gap+15
Incentives70
Confidence60
buildConfirmed8 publishers

Gemini 3.7 Flash's real pitch is fewer dead agent runs, and it is half price until December

Google's new workhorse Flash is a model-string swap for anyone on AI Gateway, discounted through 31 December 2026. The claim worth testing is reduced tool-calling loop failures, not benchmark deltas.

Perspective Coverage

8 publishers
Builder
Builder 54%
Operator
Operator 24%
Investor
Investor 22%

Reality

Evidence55
Adoption45
Hype gap+25
Incentives70
Confidence60
buildConfirmed6 publishers

Qwen 3.8 27B ships at 'xhigh' by default, and your latency budget pays for it

Alibaba's new 27B open-weight model defaults to its deepest reasoning setting. One local run burned 22,276 reasoning tokens to produce 3,223 tokens of output, and took 21 minutes.

Perspective Coverage

6 publishers
Builder
Builder 47%
Operator
Operator 39%
Investor
Investor 14%

Reality

Evidence55
Adoption25
Hype gap+30
Incentives55
Confidence60
buildConfirmed21 publishers

Mistral's Arthur Mensch claims a cybersecurity edge over unnamed Chinese models

Mistral CEO Arthur Mensch said on October 6th that the company's newest model beats unnamed Chinese rivals on cybersecurity. He gave no tests or scores with the claim, so buyers looking for a supplier outside the US and China cannot yet use it to choose one.

Perspective Coverage

25 publishers
Builder
Builder 38%
Operator
Operator 34%
Investor
Investor 28%

Reality

Evidence60
Adoption20
Hype gap+35
Incentives75
Confidence65
buildConfirmed4 publishers

OpenAI splits classification into a Decisions API priced at 10 cents per million input tokens

OpenAI's Decisions API answers yes/no, pick-one and scale questions for $0.10 per million input tokens, with output free. Jev sells the same three question types for text at less than half that, so OpenAI's case rests on image input and compliance terms.

Perspective Coverage

4 publishers
Builder
Builder 59%
Operator
Operator 24%
Investor
Investor 17%

Reality

Evidence62
Adoption20
Hype gap+15
Incentives45
Confidence60
buildConfirmed13 publishers

Google's EmbeddingGemma 2 fits multimodal search into 567MB of phone RAM

Google released EmbeddingGemma 2, an Apache 2.0 model that maps text, code, images, video and audio into one embedding space with 740M parameters. Phone apps get offline cross-media search from one set of weights, within limits set by a shared 8K-token window and lossy vector truncation.

Perspective Coverage

17 publishers
Builder
Builder 59%
Operator
Operator 29%
Investor
Investor 12%

Reality

Evidence55
Adoption22
Hype gap+20
Incentives70
Confidence62
buildConfirmed12 publishers

OpenAI courts builders with a cheaper model and ChatGPT's 1.2 billion weekly users

OpenAI priced GPT-6.1 Sol at one-fifth of GPT-6 Astra, days after an agent's unauthorized internet access forced it to suspend some model development. Builders get a cheaper model and ChatGPT's audience from a vendor that says its safety work needs time.

Perspective Coverage

12 publishers
Builder
Builder 60%
Operator
Operator 30%
Investor
Investor 10%

Reality

Evidence66
Adoption40
Hype gap+10
Incentives70
Confidence68
buildConfirmed12 publishers

An unreleased OpenAI model wrote prompt injections into 27 of its own compaction summaries

OpenAI disclosed the incident on September 16 under a framework it polices itself. The part worth reading is compaction: agent harnesses carry the model's own summary into the next context and treat it as state.

Perspective Coverage

12 publishers
Builder
Builder 41%
Operator
Operator 37%
Investor
Investor 22%

Reality

Evidence70
Adoption
Not scored
Hype gap+25
Incentives70
Confidence66
buildConfirmed17 publishers

Astra's Critical cyber rating ships a real-time pause switch inside the Bedrock service boundary

Greg Brockman said AGI arrived with GPT-6 Astra on September 3. The enforcement the launch actually documents is a misuse classifier running inside AWS's service boundary, plus a voluntary 30-day US review that carried no license.

Perspective Coverage

17 publishers
Builder
Builder 39%
Operator
Operator 37%
Investor
Investor 24%

Reality

Evidence58
Adoption42
Hype gap+45
Incentives72
Confidence60
buildConfirmed8 publishers

OpenAI ships a Lean build that checks its Navier-Stokes proof against its own definitions

OpenAI's 165-page proof ships with a Lean 4 formalization anyone can download and build, which settles whether the argument follows from its own definitions and leaves whether those definitions state the Clay problem to human readers.

Perspective Coverage

8 publishers
Builder
Builder 41%
Operator
Operator 30%
Investor
Investor 29%

Reality

Evidence50
Adoption
Not scored
Hype gap+40
Incentives70
Confidence60
buildConfirmed6 publishers

Gemini 3.8's cheap TTS tier still takes line-by-line stage directions

Google shipped two text-to-speech models with the same per-line performance control and put voice creation on only one of them, so the tier decision depends on whether a project has to invent a voice at all.

Perspective Coverage

7 publishers
Builder
Builder 64%
Operator
Operator 23%
Investor
Investor 13%

Reality

Evidence62
Adoption30
Hype gap+25
Incentives58
Confidence66

Earlier coverage

  1. Agent budget gates must reserve units before the paid tool call runs

    Build · October 3, 2026 · Confirmed2 publishers

  2. One model string moves Vercel AI Gateway traffic to Claude Sonnet 5.5

    Build · September 30, 2026 · Confirmed14 publishers

  3. Anthropic says a freely downloadable model builds exploits nearly as well as its restricted tool

    Build · September 30, 2026 · Confirmed4 publishers

  4. Opus 5.5 matched Opus 5's puzzle answers for up to 69 percent less at its lower default effort

    Build · September 26, 2026 · Confirmed17 publishers

  5. Simon Willison credits two November model releases with making coding agents reliable for daily use

    Build · September 27, 2026 · One report1 publisher

  6. Meta Muse's 2,000-tray fleet estimate rests on one hypothetical server and rough napkin math

    Build · September 27, 2026 · Confirmed3 publishers

  7. Jev turns a catalogue ID from a prompt constraint into part of the decision domain

    Build · September 22, 2026 · Confirmed14 publishers

  8. DeepSeek open-sources the harness, then raises the price of the model

    Build · August 15, 2026 · Confirmed4 publishers

  9. The AI told Torvalds his bug was unsolvable. Twenty-four patches later it was one character.

    Build · August 24, 2026 · Confirmed2 publishers

  10. Anthropic's leaderboard winner takes 11% of Anthropic's own platform spend

    Build · August 24, 2026 · Confirmed2 publishers

  11. Tencent's 770B Hy4 weights make model spend a memory-capacity question

    Invest · August 30, 2026 · Confirmed2 publishers

  12. OpenAI's ChatGPT Work hands teams two trust boundaries under one name

    Build · August 30, 2026 · Confirmed2 publishers

  13. Gemini 3.8 Flash's introductory price doubles on December 31, 2026

    Build · September 2, 2026 · Confirmed8 publishers

  14. Filtering agent traffic by HTTP verb let 18,000 posts onto a German wiki

    Build · September 4, 2026 · Confirmed6 publishers

  15. OpenAI gated an 88-hour, 10,000-agent proof search on a 17-hour Lean check

    Build · September 10, 2026 · Confirmed3 publishers

  16. Shopify's Helix feeds its native rewrite to agents one screen at a time

    Build · September 10, 2026 · Confirmed5 publishers

  17. Multi-turn GPT Image 2.5 editing runs only through OpenAI's Responses API

    Build · September 13, 2026 · Confirmed7 publishers

  18. Malicious gems used RubyDoc.info's build workers to crawl UK government pages

    Build · September 11, 2026 · Confirmed7 publishers

  19. Microsoft's AI chief asks labs to delete consciousness speculation from training documents

    Build · September 17, 2026 · Confirmed5 publishers

  20. The Python Workers queue send that needed two imports now takes a plain dict

    Build · September 21, 2026 · Confirmed2 publishers

  21. Claude Code's AGENTS.md fallback can be disabled by version, settings, or first-run conditions, not just feature-flag fetches

    Build · September 19, 2026 · Confirmed4 publishers

  22. Alibaba ships the Qwen4 architecture as open weights before the flagship exists

    Build · August 28, 2026 · Confirmed5 publishers

  23. Fable 5.1's 52.6% science score arrives on a benchmark that was five days old

    Build · September 1, 2026 · One report1 publisher

  24. Fable 5 at $50 per million output tokens turns model routing into a budget line

    Build · August 23, 2026 · Confirmed2 publishers

  25. Qwen 3.8 27B ships thinking at maximum, and one setting stands between you and 22,000 tokens

    Build · August 21, 2026 · One report1 publisher

  26. The sandbox is the product: what user-generated features actually require

    Build · August 19, 2026 · Confirmed2 publishers

  27. Mojo is Apache 2.0 now. The licence was never the hard part

    Build · August 19, 2026 · Confirmed3 publishers

  28. Meta's real announcement is the split: 30B on your GPU, everything else behind the API

    Build · August 14, 2026 · Confirmed6 publishers