iiinova — where your digital self keeps evolving

OpenAI DevDay 2026 live blog

I'm at OpenAI's DevDay event in San Francisco today - as I have for the past three DevDay events, I'm running a live blog where I'll be posting updates during the keynote, which starts in five minutes simonwillison.net/2026/Sep/29/...

simonwillison.net

I’m at OpenAI DevDay today, in Fort Mason, San Francisco. Same as last year I’ll be live blogging the keynote and some other notes during the day. OpenAI gave me …

48811

Anthropic told investors that their Q2 was profitable, and it looks like their Q3 is going to be profitable too They haven't shared audited financials yet, I expect those will come with their S-1

030

I don't think we have a Fable-class open weight model yet - wouldn't be surprised if we got there by the end of the year though

041

Maybe if they can convince every other lab to slow down too!

250

No idea, I think the fact that the Reuters story only includes 2025 numbers is really weird

190

I'd love to know that We can at least estimate based on how much it costs to run similarly capable open weight models - Kiki K3 etc

170

I'm pretty skeptical of those trillion dollar infrastructure numbers - they sound very bubbly to me Billions of dollars I can believe - Anthropic are already renting environmentally nasty data centers from SpaceX for over a billion dollars a month, and that's because they needed the extra capacity

050

They're rumored to make a healthy margin on inference - so the more tokens they can sell the better Especially since corporate clients can't sign up for the $100 or $200/month subscription plans - companies have to pay the full API prices

3170

They've caught up a bit in the past couple of months with Codex Desktop and GPT-5.6 and GPT-6 - but enterprise companies like signing longer deals and Anthropic snapped up a whole lot of those in the first half of the year

030

OpenAI do seem a whole lot more enthusiastic than Anthropic at shouting about spending a trillion dollars on data centers or whatever - I think they blow a lot more hot air (which is saying a lot)

040

OpenAI were blowing untold amounts of money on distractions like Sora - then in April they refocused because Anthropic were eating their lunch, and now they're trying to run the same playbook

240

It was coding agents. Prior to Claude Code and Codex (and the Opus 4.5 / GPT 5.1 models that made those actually work) it was hard to spend more than $50/month on AI tokens, there just wasn't enough useful stuff to do with them A good coding agent can justify $50/day or more

5201

Last year I didn't think OpenAI or Anthropic were long-term sustainable at all I thought their plan was to should "AGI! AGI!" at investors, keep the funding rolling in, and desperately hope to find a business model This year they found a business model, and it's not "AGI", it's coding agents

360

The 2025 ones? Sure, if they kept on losing money at that rate in 2026. But all indications are that the opposite has happened - in 2026 they found product-market fit for a product that companies are spending thousands of dollars per employee per month on They didn't have that last year

3120

Anthropic's only existed for five years, and the stories I've seen about them being financially sustainable have only been about 2026

190

I mean Uber set a cap of $1,500 per employee per AI tool they were using, and if they have ~3,000 engineers on staff that's a cool $4.5m per month from just one customer, AFTER they set those limits

340

Anthropic claimed that two of their quarters this year have been profitable - I'm holding out for the actual numbers in the full S-1 because of the many different ways they might be cooking the books, but I think there's a decent chance those claims will hold up

2120

Yes, you're right - which makes their revenue acceleration in 2026 even more impressive

450

And you know this! You know that a Reuters article about Anthropic's 2025 numbers is going to mislead people who are unaware of how much their revenue changed this year

2151

Anthropic lost $8 billion in 2025 against $12 billion in revenue... but that was before they started pulling in that much revenue /per quatter/ in 2026 as companies started blowing millions of dollars on coding agent tools that hardly existed a year ago

3181

Well yeah... your replies are filled with people who seem to think those 2025 numbers represent the current financial state of the company, when they very clearly do not

3220

Hah I did actually think of you when I thought "who do I know called Mike", I figured a more common name was less risky! (I nearly went with "Tom")

110

My hunch is that the challenge of regular non-nerds having trouble finding things to do with their agents will mostly be solved by word-of-mouth When your dumbass friend Mike figures out how to use Muse... Like how in the early days of MySpace people learned CSS to customize their page!

7320

(The full talk also covers New Zealand’s Kākāpō parrot breeding season, probably the best news of 2026)

1100

And my closing thought on how this stuff impacts our lives as software engineers, with a quote from cycling champion Greg LeMond

It doesn't get easier -
you just get faster
Greg LeMond
3x Tour de France champion

120

Here's how I define "Fable class models" - first Claude Fable 5, now Claude Opus 5.5 and GPT-Astra 6 and maybe GPT-5.6 Sol as well

Fable class models
If you can define a goal, provide unambiguous instructions,
and provide access to necessary tools They can solve your
problem with brute force

240

On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological …

I've published detailed notes and an annotated transcript to accompany the video of the keynote I gave at @wearedevelopers.bsky.social World Congress North America in San Jose on Friday - here's my rundown of everything that's happened with LLMs in 2026 so far simonwillison.net/2026/Sep/27/...

simonwillison.net

2026 in LLMs (so far)

310420

She made nesting bowls too! bsky.app/profile/simo...

030

JPEG XL looks not to have good browser support yet, but it looks like AVIF became usable around 2023 caniuse.com?search=AVIF

Screenshot of the CanIUse support charts for AVIF - Chrome, Firefox, Opera have had support since before 2022, Safari and Safari on iOS gained support mid-2022, Edge finally got there at the end of 2023

120

I just exported 69 slides from Keynote - as (reduced quality) JPG: 10.4MB, as PNG: 100MB, as WebP: 4.1MB Quality looks great, this 1920×1080 image here is 58KB

6260

How late to the party am I on WebP? I only recently started using it as my preferred format for non-photo images and it's SO MUCH smaller than PNG, and looks much better than JPG used to shrink down screenshots too Has everyone else been having a WebP party without me for years?

16751

But really the main thing is to poke around with them, try to get them to build impossible things, and see what they still can't do yet

120

Simon Willison on coding-agents

Feels a bit self-serving to say it, but my blog has a LOT of material on coding agents simonwillison.net/tags/coding-... - plus an in-progress guide that I need to invest more work in simonwillison.net/guides/agent...

simonwillison.net

252 posts tagged ‘coding-agents’. Systems where an LLM writes code which is then compiled, executed, tested or otherwise exercised by tools in a loop.

250

Plus saying "have you been following this year's Kākāpō breeding season?" is a fantastic conversation-starter

1130

I am now the owner of some VERY fine Kākāpō ceramics, and the lego set, and I have a plushy somewhere too bsky.app/profile/simo...

2180

Sirocco on Last Chance To See, followed by stumbling across the Kākāpō Recovery Programme first on Twitter and now on Bluesky, in particular @digs.bsky.social

230

Prompt and transcripts here simonwillison.net/2026/Sep/26/... - or you can load the HTML page Claude built to see the interactive version before it became a video tools.simonwillison.net/kakapo-party

tools.simonwillison.net

Kākāpō Party

1160