Gemini 3.5 Pro Shelved for Gemini 4?

On May 19, Google announced that Gemini 3.5 Pro would arrive “next month,” setting its target release for June. As of August 26, the flagship model still lacks a release date, while executive messaging has already shifted toward Gemini 4.

SemiAnalysis reported that Google quietly shelved 3.5 Pro to focus resources on its next generation. Google has denied any cancellation, maintaining that partner testing remains underway. Yet following early performance concerns in June, the two-month delay has fueled industry speculation that 3.5 Pro will be skipped entirely.

A June Promise, Still Unmet in August

Google's official announcement stated that 3.5 Pro was already in internal use and would launch the following month. June passed without a release. In July, Google launched 3.6 Flash alongside lighter models, leaving the flagship's release date at an indefinite “when ready.”

Reuters reported that internal testing showed the flagship trailing rival models in coding performance, forcing a two-month delay. Google acknowledged that coding and agentic workflows required further refinement. The setback stemmed from technical hurdles in hitting target benchmarks rather than simple scheduling adjustments.

Google Pivots Public Focus to Gemini 4

The model now filling the roadmap is not 3.5 Pro, but Gemini 4. CEO Sundar Pichai stated that competing at the frontier requires a larger foundation model, confirming that training is underway. Google described the effort as the company's “most ambitious pre-training run yet,” according to 9to5Google.

Pretraining Gemini 4 marks only the initial phase of development, not an impending release. Post-training, safety evaluations, product integration, and inference optimization must follow. Google has not provided an official timetable. Even a year-end release in November or December would leave a five-to-six-month gap from the original June target.

Gemini 3.5 Pro could still appear as a bridge release. However, Google's compute allocation and strategic focus have already shifted to Gemini 4. The underlying issue extends beyond a single delayed launch: an entire intermediate model generation has lost its standing in the roadmap.

Gemini 3.7 Flash Anchors the Value Tier

The disrupted roadmap creates an immediate challenge: what sustains the lineup until Gemini 4 arrives? Google's response was not another Pro model, but Gemini 3.7 Flash. Rather than positioning it as a frontier-reclaiming flagship, Google framed it as a workhorse model optimized for coding and agentic tasks. Pricing reinforces that role: through year-end, API pricing is set at $0.75 per million input tokens and $3.75 per million output tokens, halving the cost of 3.6 Flash.

Gemini 3.7 Flash remains competitive, scoring 56 on the Artificial Analysis Intelligence Index with high throughput and low latency. Yet its core strength lies in task-level cost efficiency rather than top-tier reasoning. In the absence of 3.5 Pro, a budget-focused offering is holding ground it was never designed to defend alone.

That value tier faces mounting competitive pressure as well. SpaceXAI's Grok 4.6 scored 61 on the latest Intelligence Index, while Chinese developer Z.ai's open-weight GLM-5.3 reached 60. In the latest comparison, Gemini 3.7 Flash remains at 56. Google's current offering trails both a proprietary US flagship and an open-weight Chinese release.

Gemini 4 Must Prove Itself Amid Key Departures

Beyond benchmark results, Google faces leadership transitions and talent loss across its AI division. An August reorganization shifted Demis Hassabis from operational management to board chair at DeepMind, while early Gemini co-leads Jeff Dean and Oriol Vinyals departed to launch a new AI startup.

Earlier departures included co-lead Noam Shazeer to OpenAI and AlphaFold researcher John Jumper to Anthropic. As the DeepMind reorganization reported by Reuters indicates, the loss of foundational architects creates substantial execution risk for Gemini 4.

SemiAnalysis argued that DeepMind is no longer a frontier lab and that Gemini 4 will struggle to reverse current momentum. While that represents an aggressive analyst perspective, Google still commands custom TPU infrastructure, massive capital reserves, and distribution reach spanning Search, Android, and Workspace.

Even so, Gemini 4 must outperform rival models at its eventual launch rather than where the market stood when pretraining began. The company must navigate compute allocation, research turnover, and execution timelines. Whether the Gemini 3.5 Pro delay marks a temporary setback or a deeper strategic retreat now rests on Gemini 4.