That drastic reset came after early testers flagged structural failures in core capabilities, including gaps in math, reasoning, and recursive tool calling . A rebuilt version then missed not one, but three consecutive deadlines: its initial June target, then July 17, and then any date beyond . The rebuilt model was still plagued by hallucinations and inconsistent outputs that prevented it from clearing basic reliability standards .

SemiAnalysis's August 7 report crystallized the outcome: "Since our institutional note, Google has silently canceled Gemini 3.5 Pro" . Google has never publicly confirmed a launch date for the model on its official deprecation page—unlike its other publicly available models .

The Evidence for Quiet Cancellation

Several converging signals support the claim that Gemini 3.5 Pro has been canceled rather than simply delayed:

• SemiAnalysis's direct statement: The report states the cancellation plainly .
• Scrapped architecture: Google destroyed the original base model and started over, a move multiple sources describe as a remarkable sign of crisis .
• Consecutive missed deadlines: Three launch dates passed without a release .
• Persistent quality issues: The rebuilt version failed to fix hallucinations and structural failures in recursive tool calling .
• Market reaction: Alphabet lost roughly $200 billion in market value on reports that Gemini 3.5 Pro was months behind schedule .
• No official page: Google's deprecation/status page lists shipping dates for Flash models but contains no entry or release date for 3.5 Pro .

It is important to note that Google has not issued a public statement confirming the cancellation. The company's live Gemini page still says "3.5 Pro coming soon," and in its July 21 announcement, Google stated the model was "currently testing with partners and we plan to make it broadly available as soon as it's ready" .

The Stopgap: Gemini 3.6 Flash and Other Lighter Models

On July 21, 2026, Google shipped three new lightweight models, none of which was the flagship Pro :

• Gemini 3.6 Flash — positioned as a bridge model
• Gemini 3.5 Flash-Lite — a smaller, cheaper variant
• Gemini 3.5 Flash Cyber — likely a security-focused variant

The release was widely interpreted as a stopgap measure. SemiAnalysis offered a blunt assessment of Gemini 3.6 Flash: it is "generally worse than Muse Spark 1.2, Grok 4.5, and the tier 1 Chinese open-source models" . As a result, the report ranks Gemini 8th or 9th overall in the current large-model landscape, depending on how you count .

DeepMind's Leadership Exodus

On August 5, 2026—just two days before the SemiAnalysis report—Google announced a sweeping DeepMind leadership overhaul that further underscored the turmoil :

• Demis Hassabis stepped away from day-to-day operations as DeepMind CEO.
• Jeff Dean (Google Chief Scientist and Gemini co-lead) left to start a new lab called Discovery Loop, taking with him Sanjay Ghemawat, Quoc Le, and Oriol Vinyals (another Gemini co-lead).
• Koray Kavukcuoglu, the former DeepMind CTO and the one remaining Gemini co-lead, replaced Hassabis as leader of DeepMind/Gemini.

SemiAnalysis's commentary on the timing: "Obviously these are not the actions of people excited about Gemini 4 Pro" .

Outlook for Gemini 4

Google has started pretraining Gemini 4 and has publicly hyped it as the next flagship model . But the SemiAnalysis report is deeply skeptical that the new model can turn things around. The core problem identified by the analysts is not technology alone but a fundamental lack of conviction at the corporate level .

The report points to a specific conflict: Google Cloud (GCP) has sold enormous amounts of TPU compute to AI competitors. Over 20% of TPU shipments from the third quarter of 2026 through the fourth quarter of 2027 are going to Anthropic, on long-term contracts with no clawback provision for DeepMind . This means Google is simultaneously funding its own AI rival and limiting DeepMind's access to its own hardware.

Combined with the steady exodus of top AI talent—the latest being Jeff Dean and a chunk of Gemini's leadership—SemiAnalysis concludes bluntly: "We believe DeepMind is no longer a frontier lab" and "their odds of reaching SOTA again have dropped to zero" .

Whether Gemini 4 can reverse that verdict remains to be seen. What is clear is that the road to it has been paved with the quiet cancellation of what was supposed to be Google's most important AI release of 2026.

That drastic reset came after early testers flagged structural failures in core capabilities, including gaps in math, reasoning, and recursive tool calling . A rebuilt version then missed not one, but three consecutive deadlines: its initial June target, then July 17, and then any date beyond . The rebuilt model was still plagued by hallucinations and inconsistent outputs that prevented it from clearing basic reliability standards .

SemiAnalysis's August 7 report crystallized the outcome: "Since our institutional note, Google has silently canceled Gemini 3.5 Pro" . Google has never publicly confirmed a launch date for the model on its official deprecation page—unlike its other publicly available models .

The Evidence for Quiet Cancellation

Several converging signals support the claim that Gemini 3.5 Pro has been canceled rather than simply delayed:

• SemiAnalysis's direct statement: The report states the cancellation plainly .
• Scrapped architecture: Google destroyed the original base model and started over, a move multiple sources describe as a remarkable sign of crisis .
• Consecutive missed deadlines: Three launch dates passed without a release .
• Persistent quality issues: The rebuilt version failed to fix hallucinations and structural failures in recursive tool calling .
• Market reaction: Alphabet lost roughly $200 billion in market value on reports that Gemini 3.5 Pro was months behind schedule .
• No official page: Google's deprecation/status page lists shipping dates for Flash models but contains no entry or release date for 3.5 Pro .

It is important to note that Google has not issued a public statement confirming the cancellation. The company's live Gemini page still says "3.5 Pro coming soon," and in its July 21 announcement, Google stated the model was "currently testing with partners and we plan to make it broadly available as soon as it's ready" .

The Stopgap: Gemini 3.6 Flash and Other Lighter Models

On July 21, 2026, Google shipped three new lightweight models, none of which was the flagship Pro :

• Gemini 3.6 Flash — positioned as a bridge model
• Gemini 3.5 Flash-Lite — a smaller, cheaper variant
• Gemini 3.5 Flash Cyber — likely a security-focused variant

The release was widely interpreted as a stopgap measure. SemiAnalysis offered a blunt assessment of Gemini 3.6 Flash: it is "generally worse than Muse Spark 1.2, Grok 4.5, and the tier 1 Chinese open-source models" . As a result, the report ranks Gemini 8th or 9th overall in the current large-model landscape, depending on how you count .

DeepMind's Leadership Exodus

On August 5, 2026—just two days before the SemiAnalysis report—Google announced a sweeping DeepMind leadership overhaul that further underscored the turmoil :

• Demis Hassabis stepped away from day-to-day operations as DeepMind CEO.
• Jeff Dean (Google Chief Scientist and Gemini co-lead) left to start a new lab called Discovery Loop, taking with him Sanjay Ghemawat, Quoc Le, and Oriol Vinyals (another Gemini co-lead).
• Koray Kavukcuoglu, the former DeepMind CTO and the one remaining Gemini co-lead, replaced Hassabis as leader of DeepMind/Gemini.

SemiAnalysis's commentary on the timing: "Obviously these are not the actions of people excited about Gemini 4 Pro" .

Outlook for Gemini 4

Google has started pretraining Gemini 4 and has publicly hyped it as the next flagship model . But the SemiAnalysis report is deeply skeptical that the new model can turn things around. The core problem identified by the analysts is not technology alone but a fundamental lack of conviction at the corporate level .

The report points to a specific conflict: Google Cloud (GCP) has sold enormous amounts of TPU compute to AI competitors. Over 20% of TPU shipments from the third quarter of 2026 through the fourth quarter of 2027 are going to Anthropic, on long-term contracts with no clawback provision for DeepMind . This means Google is simultaneously funding its own AI rival and limiting DeepMind's access to its own hardware.

Combined with the steady exodus of top AI talent—the latest being Jeff Dean and a chunk of Gemini's leadership—SemiAnalysis concludes bluntly: "We believe DeepMind is no longer a frontier lab" and "their odds of reaching SOTA again have dropped to zero" .

Whether Gemini 4 can reverse that verdict remains to be seen. What is clear is that the road to it has been paved with the quiet cancellation of what was supposed to be Google's most important AI release of 2026.