Google’s next-generation flagship mannequin, Gemini 3.5 Professional, has missed its third consecutive launch deadline, extending a delay that started shortly after the mannequin was first introduced at Google I/O in June 2026. Rather than the flagship, Google has begun rolling out a collection of smaller, sooner interim fashions — Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a security-focused variant referred to as Gemini 3.5 Flash Cyber — an method that retains the corporate visibly energetic out there whereas the core Professional mannequin continues its prolonged improvement cycle.
A Sample of Slipping Deadlines
Gemini 3.5 Professional was initially positioned at I/O because the mannequin that will restore Google’s standing on the frontier of AI functionality. That authentic June goal got here and went and not using a launch. A subsequent goal of July 17 additionally handed and not using a public launch. In response to trade reporting, the underlying trigger just isn’t merely a scheduling situation: after the preliminary June delay, Google DeepMind reportedly ordered a full ground-up pre-training restart for the mannequin, quite than trying to patch or fine-tune the present checkpoint. Sources aware of the interior testing course of have pointed to unsatisfactory code-generation efficiency throughout closed analysis because the direct set off for that restart, alongside persistent points with hallucination charges and inconsistent outputs that fell in need of what Google thought-about production-ready.
A ground-up restart of that sort is a big determination for any lab to make this late in a mannequin’s introduced improvement cycle, and it suggests the issues uncovered internally went deeper than what incremental fine-tuning may fairly repair within the time out there.
The Expertise Query
The delays have coincided with a separate, and arguably extra structurally regarding, sign: a wave of senior departures from the Gemini analysis staff. Within the week of June 21 by 27, 2026, 4 senior Gemini researchers introduced they have been leaving Google, with locations reportedly together with Anthropic. That departure wave was not an remoted occasion; it continues a broader sample seen throughout 2025 and 2026 during which Google has misplaced a gradual stream of senior AI researchers to Anthropic, OpenAI, and a variety of well-funded startups.
A scheduling delay is, in precept, a solvable engineering drawback. A sample of senior researchers selecting to depart in the midst of a high-stakes mannequin rebuild is a distinct sort of sign, and one that’s tougher to repair on any predictable timeline — notably when the locations these researchers are selecting embrace the corporate’s most direct competitor.
The Stopgap Technique
To keep away from an prolonged silence on new releases whereas the Professional rebuild continues, Google has turned to a well-known trade playbook: delivery smaller, sooner fashions to maintain its API ecosystem energetic and provides builders one thing present to construct in opposition to. On July 21, 2026, Google DeepMind rolled out Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber throughout AI Studio, Vertex AI, and restricted early-access shopper surfaces.
Gemini 3.6 Flash is priced at $1.50 per million enter tokens and $7.50 per million output tokens, and Google says it cuts output pricing by roughly seventeen % in contrast with the prior Flash technology whereas additionally decreasing typical token consumption by an identical margin. Gemini 3.5 Flash-Lite targets high-volume, latency-sensitive workloads at a steep low cost, priced at simply $0.30 per million enter tokens and $2.50 per million output tokens. The unique Gemini 3.5 Flash, which launched again on Could 19, has in the meantime continued carrying the majority of Google’s manufacturing site visitors all through the Professional delay.
Early unbiased testing of the brand new Flash fashions has been blended. On Synthetic Evaluation’s composite metrics, Gemini 3.6 Flash reportedly achieved solely tough parity with its predecessor quite than a transparent enchancment, and on CursorBench, an analysis centered on coding-assistant efficiency, it reportedly fell behind Cursor’s personal native mannequin. Developer suggestions within the days following launch pointed to particular weaknesses in frontend code technology and spatial reasoning, regardless of comparatively robust official benchmark figures — a niche that a number of reviewers have attributed to a well-known sample in AI releases: benchmark optimization doesn’t at all times translate cleanly into real-world, production-grade usability.
Markets Are Watching Intently
The uncertainty round Gemini 3.5 Professional’s precise launch date has change into concrete sufficient to indicate up in prediction markets. On Polymarket, a market monitoring when Google will launch a brand new Gemini reasoning flagship has accrued greater than $323,000 in buying and selling quantity, with the main consequence at present putting an eighty-one % chance on a July 31 supply date. A separate market monitoring the subsequent Gemini Professional launch particularly exhibits August 7 as the present favored consequence, at seventy-three %. As a result of these are markets the place members are staking actual cash quite than merely expressing an opinion in a ballot, the shifting odds are a fairly direct sign that merchants more and more doubt both of Google’s two already-missed dates shall be met a 3rd time round, and are pricing in an additional slip.
A Widening Aggressive Hole
Each week that Gemini 3.5 Professional stays unreleased is every week during which Google’s rivals proceed to ship. For the reason that authentic June delay, Anthropic has launched Claude Fable 5 — briefly pulled from public entry over export-control issues earlier than being restored on July 1 — adopted by Claude Sonnet 5 and, most not too long ago, Claude Opus 5. OpenAI’s GPT-5.6 Sol has been in energetic use all through the identical window, and Moonshot AI’s Kimi K3 has simply change into the biggest open-weight mannequin ever launched, with robust showings on unbiased coding benchmarks. That leaves Google as the one main frontier AI lab and not using a 2026 flagship mannequin normally manufacturing on the date its personal management initially promised.
Enterprise groups at present evaluating frontier AI platforms for the second half of the yr are, in follow, selecting amongst GPT-5.6 Sol, Claude Opus 5 or Fable 5, Grok 4.5, and now Kimi K3 — and with every further week that passes, Gemini’s flagship tier is just absent from that comparability solely, quite than merely trailing behind on particular benchmarks.
What Comes Subsequent
In response to reporting on Google’s inside planning, one choice into consideration is skipping the Gemini 3.5 Professional technology altogether and redirecting sources towards Gemini 4.0, with an emphasis on sensible, real-world reliability over chasing incremental benchmark positive aspects — a lesson the corporate seems to be drawing instantly from the three.5 Professional rebuild’s struggles. That might be an uncommon step for a corporation that has already introduced the mannequin publicly and hooked up a CEO-level timeline to it, however it might not be unprecedented in an trade the place a product arriving many months late can successfully belong to a distinct aggressive period than the one it was initially designed for.
For now, Google’s public technique is to maintain delivery what it will possibly — sooner, cheaper Flash-tier fashions — whereas the flagship rebuild continues out of public view. Whether or not that method is sufficient to maintain developer mindshare till Gemini 3.5 Professional, or no matter finally replaces it, lastly ships stays one of many extra carefully watched open questions in AI heading into the again half of 2026.
Studying the Delay in Context
A single missed deadline hardly ever tells you a lot about an organization’s underlying technical place; launch dates slip for every kind of odd causes, from last-minute security opinions to infrastructure bottlenecks that don’t have anything to do with the mannequin’s precise high quality. Three consecutive misses, paired with a full pre-training restart and a wave of senior departures, is a distinct sort of sign. It suggests the issue Google is having with Gemini 3.5 Professional just isn’t a matter of polish or closing tuning, however one thing nearer to the mannequin not but assembly the bar Google itself set earlier than it was prepared to place the “Professional” label on it and ship it to the identical enterprise clients at present evaluating it in opposition to Claude Opus 5 and GPT-5.6 Sol.
That warning is, in a slim sense, defensible. Delivery a flagship mannequin that underperforms on arrival can do extra lasting injury to an organization’s aggressive place than delivery late, notably in a market the place enterprise consumers are actively operating head-to-head comparisons throughout a number of distributors directly. The danger Google is taking by ready is a distinct one: that by the point Gemini 3.5 Professional, or its eventual alternative, does ship, the aggressive bar it must clear may have moved even additional out of attain, given how shortly Anthropic, OpenAI, and now Moonshot AI have continued to launch new fashions all through the precise interval Google has spent rebuilding.
The Stakes for Google’s Broader AI Technique
Gemini’s flagship tier issues to Google properly past the standalone chatbot market. It underpins the AI options constructed into Google Search, Workspace, and Android itself, and it’s central to how Google positions itself in opposition to rivals in cloud AI infrastructure by Vertex AI. A chronic hole on the flagship tier doesn’t simply value Google standing in head-to-head mannequin comparisons; it slows the tempo at which enhancements can propagate by each different Google product that is determined by the underlying mannequin. That’s seemingly a part of why the corporate has prioritized maintaining one thing present in market, even an interim Flash launch with acknowledged weaknesses, quite than going totally darkish till the Professional rebuild is full.
Leave a Reply