Home/Archive/13 July 2026
Weekend digest · 10–13 July 2026

The weekend the blue links finally went dark

A quiet weekend on the release front. The industry's big guns all fired on 8 and 9 July (GPT-5.6, GPT-Live, Grok 4.5 and Mistral's Robostral Navigate all landed then, just outside this issue's window), and most labs went silent through the weekend. The stories below are the ones that actually broke between Friday morning and Monday morning, and the most useful one for Claude subscribers is Anthropic extending included Fable 5 access for another week.

Google (DeepMind)

Google Search now runs entirely on Gemini 3.5 Flash

If you noticed Google results looking different this weekend, it was not your imagination. As of 10 July, every Google Search result is a Gemini 3.5 Flash generated summary, with source links embedded inside the AI page rather than listed as the familiar ten blue links. This completes the rollout Google announced at I/O in May, and it matters to anyone who publishes on the web or relies on organic search traffic: the default experience of the world's most visited website is now an AI answer rather than a list of destinations.

If you run a website, your click-through assumptions from even a month ago no longer hold, and it is worth checking how your pages are being cited inside the new format. Worth noting alongside this: community trackers place Gemini 3.5 Pro's general availability at 17 July, though Google has not confirmed that date.

Sources
Build Fast with AI, 11 Jul Endroid, Jul
Anthropic

Fable 5 stays included on paid plans until 19 July

If you are on a Claude Pro, Max, Team or eligible Enterprise plan, you have another week of Fable 5 at no extra cost. Anthropic announced over the weekend that included access, previously due to end on 12 July, now runs until 19 July, with usage capped at half of your weekly limits. This is the second extension in a week, and it is widely read as a response to OpenAI shipping GPT-5.6 in the same window.

Rather than pushing subscribers onto prepaid credits (10 dollars per million input tokens and 50 dollars per million output tokens), Anthropic reset weekly limits and gave everyone seven more days with its most capable model. If you have heavy Fable 5 workloads, front-load them this week: after the 19th the meter starts running. Anthropic says it wants to restore Fable 5 as a standard subscription offering when capacity allows, but has given no timeline.

Sources
BleepingComputer, 12 Jul TechTimes, 12 Jul
Meta

Meta's custom Iris AI chip enters production in September

Reports on 10 July confirmed that Meta will start manufacturing Iris, its custom AI processor built with Broadcom and fabricated by TSMC, in September. The chip cleared initial bug testing in around six weeks, which is unusually quick for silicon of this complexity, and it is intended to complement rather than replace the GPUs Meta buys from Nvidia and AMD.

For AI users the relevance is indirect but real. Meta plans 7 gigawatts of compute this year and 14 gigawatts next year, and custom silicon is how it plans to afford that. More cheap inference capacity at Meta means faster and more capable AI features across WhatsApp, Instagram and the Meta AI assistant, plus more pressure on rivals to keep their own inference prices down.

Sources
The Motley Fool, 10 Jul MLQ News, Jul
Hugging Face
Also noted · Hugging Face [Direct]

The only official Hugging Face publication in the window is Part 3 of its PyTorch profiling series, covering how to profile attention layers, published on 10 July on the company blog. Useful if you are optimising your own training or inference code, but not a product release. Hugging Face Blog, 10 Jul →

Quiet this weekend
OpenAI The GPT-5.6 family and GPT-Live both shipped on 8 and 9 July, just before the cutoff, and official channels were quiet all weekend. The Apple lawsuit filed on 11 July is a legal matter with no impact on product availability today.
xAI · Mistral Grok 4.5 and Robostral Navigate both launched on 8 July, outside the window, and both firms went quiet over the weekend.
DeepSeek · Qwen (Alibaba) Nothing significant from either lab. Alibaba delisted several older DeepSeek models from Model Studio on 9 July, recommending Qwen 3.7 replacements, but that predates the window.
Microsoft · Nvidia A leaked prototype of an agent-centric OS codenamed Aion circulated over the weekend, but leaks are excluded here. SK Hynix's record Nasdaq debut on 10 July renewed coverage of its Nvidia memory partnership, but that is a financial story, not a product one.
IBM · Snowflake Nothing significant in the last 72 hours from either company.

The weekend in three lines

Distribution was the story, not models. The most consequential change for ordinary users was not a new model but Google quietly finishing the switch of its entire Search product to Gemini 3.5 Flash, which changes how billions of people encounter the web every day. The custom silicon race is accelerating too, with Meta's Iris chip heading to production in September as the labs try to escape GPU economics: expect inference prices to keep falling as that capacity lands. And the frontier price war is now being fought with access windows, not just benchmarks. Anthropic's second Fable 5 extension in a week, landing days after GPT-5.6 shipped, shows the labs competing on who lets you use their best model for longer. The release calendar is bunching: nearly everything notable this cycle shipped in a 48-hour burst on 8 and 9 July, and if the rumoured 17 July Gemini 3.5 Pro date holds, this week will be another loud one.

← 10 July 2026 All issues →

Every day I sift through the noise, cut out the hype, and serve up the AI updates that actually matter for your business. Straight to your inbox before 8am. No fluff, no jargon, no faff.

👩👨👩👨+
Join 2,400+ business professionals already subscribed

You're in! First issue lands tomorrow morning.