AI Race Dispatch · Nº 01 · August 2026
Google has reportedly cancelled Gemini 3.5 Pro — the model meant to carry its AI ambitions through 2026 — after months of delays and internal obstacles. In its place: an all-in bet on Gemini 4. The report comes from a single source. Google has said nothing. If it’s true, the most consequential pivot in the company’s AI history is already under way.
Read the storyRead this first
This story rests on one report, relayed by one outlet, unanswered by Google. That doesn’t make it false — it makes it unconfirmed. So we’ve separated every claim on this page by how solid it is. Use the filters; nothing here is hiding behind certainty it hasn’t earned.
On August 10, 2026, Geeky Gadgets published a story by Julian Horsey citing the YouTube channel Universe of AI, claiming Google has discontinued Gemini 3.5 Pro. The publication, the attribution and the claim are all real and checkable.
As of publication, Google has issued no official statement confirming or denying the cancellation. No press release, no blog post, no executive comment. The silence itself is part of the story.
Months of published coverage tracked Gemini 3.5 Pro slipping: leaked specs, missed dates, a stopgap model floated to fill the gap. The timeline below reconstructs that trail from public reporting.
That 3.5 Pro is dead, that delays and internal obstacles killed it, and that its people and compute are being redirected to Gemini 4 — all of this comes from the Universe of AI report. One source, no corroboration yet.
That Demis Hassabis weighed stepping down before being kept on in a revised role, and that a new leader named Cory was appointed to steer the lab — again, reported, not announced.
That Meta’s Museark 1.2 and SpaceX’s Gro 4.5 now outscore the Gemini line on key benchmarks, and that Google’s position in AI performance rankings has slipped. Cited figures don’t exist publicly — treat the numbers on this page as illustrative.
That pieces of the cancelled 3.5 Pro project could resurface inside a lighter model possibly branded Gemini 3.7 Flash. This is inference layered on the report — the least solid ground on the page.
One report. No press release. No denial.
Chapter 01 The cancellation
Companies don’t cancel mid-generation flagships over minor slip-ups. According to the report, Google looked at Gemini 3.5 Pro — months late, stuck on internal obstacles, watching rivals ship — and decided the model wasn’t a product to fix. It was a distraction to cut.
The decision was reportedly made without fanfare: no announcement, no sunset post, no farewell benchmark run. Insiders describe a project that kept missing its own deadlines while the organization around it argued about direction. At some point, the calculus inside Google flipped — every engineer-month spent reviving 3.5 Pro was one not spent on Gemini 4, the release now expected to carry the entire Gemini franchise.
Read generously, it’s discipline: a company refusing to ship something mediocre into the most competitive market in technology. Read harshly, it’s an admission that Google spent a year building the wrong thing. Both readings are compatible with the same set of reported facts — which is precisely why the industry can’t decide what to think.
The product page that will never exist — reconstructed from the report’s claims.
“Cancelling a flagship is what a company does when the next one matters more than its own pride.”
— The charitable reading. The harsh one involves a wasted year.
Chapter 02 The unravelling
The cancellation didn’t arrive out of nowhere. Reconstructed from the trail of published coverage, here is how Gemini 3.5 Pro went from leaked sensation to reported casualty. Dates are approximate; every entry reflects claims in reporting, not Google statements.
c. early June 2026
Gemini 3.5 Pro surfaces in leaks touting a 2-million-token context window. Expectations snap into place: this is supposed to be the model that re-establishes Google at the front of the race.
Reportedc. late June
The first delay reports appear, citing early performance flaws in internal testing. The launch window starts to move — quietly at first.
Reportedc. early July
Coverage pins the launch to July 17 — pushed back specifically so engineers can upgrade the model’s mathematical reasoning. A fixable issue, reportedly. But the list keeps growing.
Reportedc. mid July
A new name enters the coverage: Gemini 3.6 Flash, a lighter model that may ship to fill the silence while 3.5 Pro is repaired. The flagship is now sharing its own stage.
Reportedc. late July
Reports explain the delays — and note Google’s focus shifting toward the 3.6 Flash stopgap and beyond. Inside the roadmap, 3.5 Pro is no longer the center of gravity.
Reportedc. early August
The framing changes from “delayed” to “deprioritized”: Google is reportedly favoring Gemini 4 over patching 3.5 Pro’s flaws. The successor stops waiting for its predecessor.
ReportedAug 10, 2026
Universe of AI reports the decision is final: 3.5 Pro discontinued, resources moved to Gemini 4. Geeky Gadgets carries the story the same day. Google says nothing at all.
ReportedChapter 03 The shake-up
Behind the cancelled model sits a messier human story. The report describes a leadership team that nearly broke apart mid-crisis — then reorganized itself to survive it.
Demis Hassabis — DeepMind’s co-founder, Nobel laureate, and the public face of Google’s AI ambitions — reportedly considered stepping down. He was instead retained in a revised role, a move the report links to steadying confidence in the company. Whatever the revised role contains, “the founder almost left” is not a sentence any organization wants in circulation.
Into the gap steps a new leader identified in the report only as Cory, appointed to guide the lab through the turbulence. New leadership can reset a drifting organization — or scramble workflows that were already fragile. Transitions like this are how labs find their footing. They’re also how they lose a year.
Before (reported)
Chief executive, Google DeepMind
Reportedly weighing resignation as delays mounted.
After (reported)
Retained — revised role
Kept in place, the report says, to steady the ship and the stock.
Newly appointed to guide the lab
Surname undisclosed in the report. Tasked with direction and delivery.
The race didn’t wait for Google.
Chapter 04 The competition
While Gemini 3.5 Pro slipped, the report says the frontier moved on without it. Meta’s Museark 1.2 and SpaceX’s Gro 4.5 are cited as now outperforming the Gemini series on key benchmarks — and Google’s position in AI performance rankings has declined accordingly.
Illustrative figures. The report claims rivals now lead on key benchmarks but publishes no verified numbers; scores below visualize the reported ordering, not measured results.
The names matter less than the direction. Meta and SpaceX attacking the frontier from opposite instincts — open-weight ubiquity versus brute-force scale — reportedly share one property: their newest models beat Google’s on the benchmarks the industry watches. In a market where perception compounds, every month spent un-shipping 3.5 Pro was a month rivals spent converting Google’s lead into folklore.
Chapter 05 Inside the lab
The most damaging claims in the report aren’t about any model. They’re about the organization. Former employees and industry insiders describe a DeepMind struggling with three interlocking failures:
Layers of process that slow decisions exactly when the market rewards speed. Work reportedly queues behind approvals while rivals ship weekly.
Factions with competing visions of what DeepMind should be — research institute, product engine, safety standard-bearer — reportedly pulling the same roadmap in different directions.
The through-line insiders keep returning to: a lab with world-class talent that can’t consistently answer what are we building, and for whom?
Shelving a high-profile project like 3.5 Pro is the kind of symptom that makes outsiders ask about the disease. Great labs have survived cancelled models. What they rarely survive is a culture that has stopped being able to finish things. The leadership shake-up is, on the report’s telling, an attempt to treat exactly that.
“DeepMind’s problem was never talent. It’s the distance between talent and a shipped product.”
— Characterizes the view attributed to former employees and insiders in the report
Chapter 06 What lies ahead
If the report is right, Google’s response runs on two tracks: salvage what worked from 3.5 Pro, and stake the franchise on what comes next.
Speculation Track one · the salvage
The speculation: pieces of the cancelled 3.5 Pro — the long-context work, the reasoning upgrades — don’t die with it. They resurface inside a lighter, faster model possibly branded Gemini 3.7 Flash, giving Google something competitive to ship while the flagship cooks. Salvage, not surrender.
No timeline. No confirmation. Pure inference from the report.
Reported Track two · the bet
The defining release. Gemini 4 inherits the reallocated engineers, the compute, and — fair or not — the verdict on this entire episode. Ship a leap forward and the cancellation becomes a story about discipline. Ship anything less, and it becomes the moment Google fell behind for good.
Expected to be the company’s next flagship. No official date.
Everything now rides on
One release to reclaim the lead —
or confirm the slide.
Chapter 07 The reaction
The industry response, per the coverage, splits into camps — and the split says as much about the reader as about Google. Move the dial.
The skeptics read the cancellation as confession. A lab that can’t ship its flagship on time, loses its leader’s full commitment, and watches rivals pass it on benchmarks doesn’t have a roadmap problem — it has an operating problem. Gemini 4, on this view, is a promise being used to cover an absence.
The middle camp shrugs at the drama. Strategic recalibration is what fast-moving markets demand; cancelling a weak release is ordinary product discipline at extraordinary scale. The only verdict that matters, they argue, arrives with Gemini 4’s benchmark numbers — everything before that is weather.
The optimists point to Google’s unmatched resources, research depth, and a long history of absorbing setbacks and re-emerging. Killing a mediocre model takes nerve; concentrating that talent and compute on one release could produce exactly the leap the rankings say Google needs.
Viewpoints paraphrased from the character of industry reaction described in the coverage — not direct quotations.
The short version
Gemini 3.5 Pro was cancelled after months of delays and internal obstacles — its people and compute redirected to Gemini 4. Google has neither confirmed nor denied it.
Demis Hassabis reportedly considered resigning and was retained in a revised role; a new leader named Cory was appointed to steer DeepMind through the storm.
Meta’s Museark 1.2 and SpaceX’s Gro 4.5 reportedly outscore the Gemini line on key benchmarks, and Google’s standing in the AI rankings has slipped.
Insiders and former employees describe structural inefficiency, internal conflict, and no cohesive direction — the conditions that turn a delay into a cancellation.
Watch for a possible Gemini 3.7 Flash salvage release — but the defining moment is Gemini 4. It decides whether this was discipline or decline.