# Why Google Rebuilt Gemini 3.5 Pro From Scratch Mid-Cycle

> DeepMind reportedly scrapped Gemini 3.5 Pro's base and retrained from scratch, delaying a leaked 17 July launch.

*Leaked reporting says DeepMind scrapped the 2.5 Pro base and ran an entirely new pretraining pass, pushing Gemini 3.5 Pro's launch to a rumored 17 July. The analyst read: a costly, revealing bet — and a strategically ironic one.*

By WireRead Editorial · WireRead
Canonical: https://wireread.com/news/gemini-3-5-pro-rebuilt-from-scratch-analysis

While OpenAI and Anthropic sprinted out of a US government review in early July, Google went the other direction: Gemini 3.5 Pro slipped. Teased at I/O in May, the model missed a June I/O target and then a 30 June general-availability date, and as of 9-10 July 2026 it remained in limited Vertex AI enterprise preview. Reporting now offers a cause — and it is the most consequential detail of the whole episode. According to leaks surfaced around 10 July, DeepMind did not merely fine-tune a late build. It abandoned the original 2.5 Pro base and ran an entirely new pretraining pass, effectively rebuilding the frontier model from scratch mid-cycle.

> **Note:** Every date and spec below comes from **leaks and unconfirmed reporting**, not from Google. The 17 July launch, the 2M-token window, the $250 tier, and the pricing are reported, not confirmed. The "below-threshold" security framing is an inference from reporting, not a Google statement.

## A rebuild is the most expensive signal a lab can send

Pretraining a frontier model is the single largest fixed cost in the business — the compute, the data pipeline, the months of wall-clock time. To scrap a base that far into a release cycle and start the run over is not a routine schedule slip; it is a deliberate decision to eat that cost twice. Labs do not absorb that lightly. The revealing part is what it implies about the alternative: DeepMind apparently judged that shipping the existing 2.5 Pro-derived build was worse than paying to retrain. That reads less like a stumble and more like a bet that the ceiling of the old base was too low to compete with what OpenAI and Anthropic were about to ship, and that only a clean pretraining run would move the frontier far enough to matter.

> As of 9-10 July 2026, Gemini 3.5 Pro remained in limited Vertex AI enterprise preview after missing its June I/O target and a 30 June GA date, with the delay attributed to DeepMind abandoning the 2.5 Pro base and running an entirely new pretraining — effectively rebuilding the model from scratch.
> — [BuildFastWithAI](https://www.buildfastwithai.com/blogs/ai-news-today-july-9-2026), 2026-07-09

## The positioning: 2M context and Deep Think at the top of the stack

The leaked specifications, if accurate, show where Google intends to compete. The headline figure is a **2-million-token context window** — reportedly the largest of any production frontier model, and a natural extension of the long-context lead Gemini has held. Reasoning is where the tiering gets pointed: **Deep Think**, the model's extended-reasoning mode, is said to be gated to the **$250/month Ultra tier**, positioning the strongest capability as a premium product rather than a default. Reported API pricing sits near **$1.25 input / $10 output per 1M tokens**.

The table below sketches the reported shape of the release. Treat every cell as leaked, not confirmed.

| Attribute | Reported detail (unconfirmed) |
| --- | --- |
| Status (9-10 July) | Limited Vertex AI enterprise preview |
| Leaked GA date | **17 July 2026** |
| Context window | **2M tokens** (reportedly largest in production) |
| Deep Think reasoning | Gated to **$250/mo** Ultra tier |
| API pricing | ~**$1.25** input / **$10** output per 1M tokens |
| US government restrictions | Reportedly **none** (below scrutiny threshold) |

Read together, these choices describe a model built to win on depth rather than breadth: a very long memory for enterprise document and codebase work, and a deliberate reasoning mode reserved for buyers willing to pay for it. The Vertex-first, enterprise-preview rollout reinforces the read — Google is aiming this build at the accounts that value context length and are least sensitive to a premium reasoning tier.

## The strategic irony: shipping freely because it scores lower

The most quietly significant claim in the reporting is a competitive one. Gemini 3.5 Pro was framed as the only major frontier model set to release without US government restrictions — reportedly because its offensive-security benchmark scores sat below the unofficial threshold that triggered scrutiny for OpenAI's GPT-5.6 and Anthropic's Fable and Mythos models. If that inference holds, the second-order effect is striking: Google's slower, later, twice-retrained model gets to ship freely precisely because it is rated less dangerous on cyber-offense. The delay and the clean-release status may share a root cause — a model that is not pushing the offensive-security frontier is both less constrained by Washington and, plausibly, why DeepMind felt it had to retrain in the first place.

> Reporting framed Gemini 3.5 Pro as the only major frontier model set to release without US government restrictions, reportedly because its offensive-security scores sat below the threshold that triggered scrutiny for OpenAI and Anthropic — with leaked details placing general availability on 17 July 2026.
> — [Fello AI](https://felloai.com/best-ai-models/), 2026-07-09

The strategic calculus, then, is subtler than "Google fell behind." A restriction-free release is a genuine go-to-market advantage — enterprise and international buyers wary of export controls or usage caveats get an unencumbered frontier model. But it is an advantage earned by scoring lower on the exact axis regulators watch, which is an awkward flag to plant. The rebuild suggests Google concluded the trade was worth it: pay to retrain, ship a cleaner, longer-context model a few weeks late, and turn the resulting compliance profile into a selling point. This story also threads off the DeepMind talent questions we covered on 25 June — a reminder that execution risk at the frontier is now as much organizational as it is technical. The tell will be the confirmed benchmarks: if a 17 July launch materializes with real numbers, we will finally see whether the second pretraining run bought Google the frontier position it paid twice for.

## Key takeaways

- As of 9-10 July 2026 Gemini 3.5 Pro was still in limited Vertex AI enterprise preview, having missed both its June I/O target and a 30 June GA date.
- Leaked details reported around 10 July attribute the slip to DeepMind abandoning the 2.5 Pro base and running an entirely new pretraining pass — effectively rebuilding the model from scratch — with GA now rumored for 17 July 2026.
- Reported (unconfirmed) specs include a 2-million-token context window, Deep Think reasoning gated to a $250/month Ultra tier, and pricing near $1.25 input / $10 output per 1M tokens.
- Reporting framed Gemini 3.5 Pro as the only major frontier model set to release without US government restrictions, reportedly because its offensive-security benchmark scores sat below the threshold that triggered scrutiny for OpenAI and Anthropic.
- None of the dates, specs, or the below-threshold inference are confirmed by Google — they are leaks and reporting, not official disclosure.

## FAQ

### Is Gemini 3.5 Pro actually launching on 17 July 2026?
That date comes from leaks reported around 10 July, not from Google. As of 9-10 July the model was still in limited Vertex AI enterprise preview after missing its June and 30 June targets, so treat 17 July as reported, not confirmed.

### What does "rebuilt from scratch" actually mean here?
Reporting attributes the delay to DeepMind abandoning the earlier 2.5 Pro base and running an entirely new pretraining pass rather than fine-tuning an existing build. Pretraining is the most expensive step in making a model, so redoing it mid-cycle is a costly, deliberate decision — and it too is reported, not officially confirmed.

### What are the leaked specs and pricing?
Leaked (unconfirmed) details include a 2-million-token context window described as the largest in any production frontier model, Deep Think reasoning gated to a $250/month Ultra tier, and API pricing near $1.25 input and $10 output per 1M tokens. No official benchmarks or confirmed pricing had been published as of 10 July.

### Why can Gemini 3.5 Pro reportedly ship without US government restrictions?
Reporting frames it as the only major frontier model set to release without those restrictions, apparently because its offensive-security benchmark scores sat below the unofficial threshold that drew scrutiny for OpenAI and Anthropic. That is an inference from reporting, not a Google statement.

### Is a restriction-free release good or bad for Google?
Both. It is a real go-to-market advantage — an unencumbered frontier model for buyers wary of usage caveats — but it is earned by scoring lower on the cyber-offense benchmark regulators watch, which is an awkward thing to advertise. The rebuild suggests Google judged the overall trade worth the extra cost.

## Sources

- [AI Updates Today (July 2026) — Latest AI Model Releases](https://llm-stats.com/llm-updates) — llm-stats, 2026-07-10
- [Best AI Models in July 2026: ChatGPT, Claude, Gemini & Grok](https://felloai.com/best-ai-models/) — Fello AI, 2026-07-09
- [AI News Today July 9 2026: 15 Biggest Stories](https://www.buildfastwithai.com/blogs/ai-news-today-july-9-2026) — BuildFastWithAI, 2026-07-09