> ## Content Index
> Fetch the complete content index at: https://www.bushletter.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Gemini 3.5 Pro stalls as coding targets missed and researchers defect
- URL: https://www.bushletter.com/gemini-3-5-pro-stalls-as-coding-targets-missed-and-researchers-defect/
- Published: 2026-07-21T04:56:49.000Z
- Updated: 2026-07-29T13:37:57.000Z
- Description: Google Cloud told developers on 19 May 2026 that Gemini 3.5 Pro was in testing and would arrive the following month. Google I/O was the stage, and the model was positioned as the company's frontier-level answer across coding, reasoning and multimodal tasks.
- Author: Editor
- Tags: Technology, US, Google DeepMind, #By Takeshi Mori

![Takeshi Mori](https://res.cloudinary.com/dz77sb7j1/image/upload/v1774262608/bushletter/authors/takeshi-mori.png)

By **Takeshi Mori** · 2026-07-21

TLDR

Google promised Gemini 3.5 Pro for June 2026, but the model was still unreleased as of late July, with internal coding benchmarks reportedly falling short of targets. Two senior figures have since departed: Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic. OpenAI shipped three new GPT-5.6 models to general availability on 9 July 2026, widening the competitive gap. The slip raises pointed questions about whether Google can hold its position at the frontier of AI development.

KEY TAKEAWAYS

01Google Cloud announced at I/O on 19 May 2026 that Gemini 3.5 Pro was in testing and due the following month.

02As of 21 July 2026, the DeepMind models page still showed '3.5 Pro coming soon' with no public release date.

03Noam Shazeer, VP of Engineering and Gemini co-lead, said on 17 June 2026 he was leaving Google to join OpenAI.

04Nobel laureate John Jumper announced on 20 June 2026 he was leaving DeepMind for Anthropic after nearly nine years.

05OpenAI released GPT-5.6 Sol, Terra and Luna models to general availability on 9 July 2026.

## What Google promised at I/O 2026

Google Cloud told developers on 19 May 2026 that Gemini 3.5 Pro was in testing and would arrive the following monthverified**Verified** Source: [cloud.google.com](https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud?ref=bushletter.com).[\[1\]](https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud?ref=bushletter.com) Google I/O was the stage, and the model was positioned as the company's frontier-level answer across coding, reasoning and multimodal tasks.

Gemini 3.5 Flash launched at the same event and showed meaningful gains over its predecessor in complex coding workflows and long-horizon tasks, but it was built for speed and cost-efficiency rather than peak performance.[\[1\]](https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud?ref=bushletter.com) The Pro variant carried Google's frontier ambitions, and the June timeline was unambiguous.

Google I/O '26 Keynote

## Why the release slipped

As of 21 July 2026, the official DeepMind Gemini models page carried a single line next to the Pro entry: "3.5 Pro coming soon"verified**Verified** Source: [deepmind.google](https://deepmind.google/models/gemini/?ref=bushletter.com), with no public release date attached.[\[2\]](https://deepmind.google/models/gemini/?ref=bushletter.com) That is more than six weeks past the promised window.

Internal benchmarks reportedly fell short of Google's own targets for coding performance, prompting engineers to keep working on the model rather than ship it. Coding performance appears to be where the gap between aspiration and reality opened up, given the Gemini 3.5 series was designed to combine advanced reasoning, multimodal understanding and agentic capabilities.[\[1\]](https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud?ref=bushletter.com) Releasing a frontier model that underperforms on benchmarks competitors have already cleared is a harder call than slipping a date.

## The talent exits that complicated the picture

Noam Shazeer, vice president of engineering and co-lead of Google's Gemini AI models, said on 17 June 2026 that he would leave Google to join OpenAI.[\[4\]](https://www.marketscreener.com/news/google-s-gemini-co-lead-noam-shazeer-to-join-ipo-bound-openai-ce7f5cdddb88fe2d?ref=bushletter.com) Shazeer said: "I'm incredibly proud of the amazing team at Google and everything we've built together."verified**Verified** Source: [marketscreener.com](https://www.marketscreener.com/news/google-s-gemini-co-lead-noam-shazeer-to-join-ipo-bound-openai-ce7f5cdddb88fe2d?ref=bushletter.com)[\[4\]](https://www.marketscreener.com/news/google-s-gemini-co-lead-noam-shazeer-to-join-ipo-bound-openai-ce7f5cdddb88fe2d?ref=bushletter.com) The departure of Gemini's co-lead, mid-delay, is not a detail that sits quietly in a footnote.

Three days later, on 20 June 2026, senior research scientist John Jumper announced he was leaving DeepMind for Anthropic.[\[3\]](https://www.linkedin.com/posts/afraemahsan%5Fjohn-jumper-leaving-google-deepmind-for-anthropic-activity-7474184647401267200-88kg?ref=bushletter.com) Jumper said: "After nearly nine years, I have decided to leave Google DeepMind and join Anthropic."[\[6\]](https://www.sahmcapital.com/news/content/us-scientist-john-jumper-to-leave-google-deepmind-for-anthropic-2026-06-19?ref=bushletter.com) Jumper is a Nobel laureate, awarded for his work on AlphaFold, and his name carries weight well beyond the AI research community.

DeepMind CEO Demis Hassabis said of Jumper's tenure: "What we achieved with AlphaFold changed the world, and showed the field what was possible with AI for science and medicine, lighting the way for how AI can benefit humanity."[\[6\]](https://www.sahmcapital.com/news/content/us-scientist-john-jumper-to-leave-google-deepmind-for-anthropic-2026-06-19?ref=bushletter.com) The statement reads more like a eulogy than a retention announcement.

## Where rivals stand while Google waits

OpenAI released GPT-5.6 Sol, Terra and Luna models to general availability on 9 July 2026, across ChatGPT, Codex and the OpenAI API.[\[5\]](https://openai.com/index/gpt-5-6/?ref=bushletter.com) Three models shipped while Google's flagship Pro release was still marked "coming soon." That is not a gap that resolves itself through press releases.

Anthropic has continued to iterate on its own frontier lineup. The competitive landscape Google mapped at I/O in May looked navigable; the one that exists in late July 2026, with a delayed model, two high-profile defections and three new OpenAI releases on the board, looks considerably different.[\[5\]](https://openai.com/index/gpt-5-6/?ref=bushletter.com)

## What the slip means for Google's model race position

Google's position in the frontier model race has always rested on the argument that research depth and compute infrastructure give it a structural advantage over newer entrants. That argument is harder to make when the researchers are leaving and the models are late.

Gemini 3.5 was supposed to demonstrate that Google could compete on peak intelligence, not just on speed and cost. A delay driven by coding benchmark shortfalls points to a specific technical problem rather than a general scheduling slip, which suggests the gap is real rather than administrative.

Developers and enterprise customers weighing Google Cloud commitments are watching the DeepMind models page as closely as any analyst. As of 21 July 2026, that page still reads "3.5 Pro coming soon," with no release date given.[\[2\]](https://deepmind.google/models/gemini/?ref=bushletter.com)

SOURCES & CITATIONS

1. [Innovations from Google I/O 26 on Google Cloud](https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud?ref=bushletter.com)
2. [DeepMind Gemini Models Page](https://deepmind.google/models/gemini/?ref=bushletter.com)
3. [John Jumper leaving Google DeepMind for Anthropic (LinkedIn)](https://www.linkedin.com/posts/afraemahsan%5Fjohn-jumper-leaving-google-deepmind-for-anthropic-activity-7474184647401267200-88kg?ref=bushletter.com)
4. [Google's Gemini co-lead Noam Shazeer to join OpenAI](https://www.marketscreener.com/news/google-s-gemini-co-lead-noam-shazeer-to-join-ipo-bound-openai-ce7f5cdddb88fe2d?ref=bushletter.com)
5. [GPT-5.6 general availability announcement](https://openai.com/index/gpt-5-6/?ref=bushletter.com)
6. [John Jumper to leave Google DeepMind for Anthropic](https://www.sahmcapital.com/news/content/us-scientist-john-jumper-to-leave-google-deepmind-for-anthropic-2026-06-19?ref=bushletter.com)

FREQUENTLY ASKED QUESTIONS

When did Google say Gemini 3.5 Pro would be released?

At Google I/O on 19 May 2026, Google Cloud announced Gemini 3.5 Pro was in testing and would be released the following month, implying June 2026.

Why has Gemini 3.5 Pro been delayed?

Internal benchmarks reportedly fell short of Google's targets for coding performance, leading to continued development rather than a public release.

Who has left Google DeepMind during this period?

Noam Shazeer, VP of Engineering and Gemini co-lead, announced on 17 June 2026 he was joining OpenAI. Nobel laureate John Jumper announced on 20 June 2026 he was leaving for Anthropic after nearly nine years at DeepMind.

What has OpenAI released while Gemini 3.5 Pro has been delayed?

OpenAI released GPT-5.6 Sol, Terra and Luna models to general availability on 9 July 2026 across ChatGPT, Codex and the OpenAI API.

![Takeshi Mori](https://res.cloudinary.com/dz77sb7j1/image/upload/v1774262608/bushletter/authors/takeshi-mori.png)

[Takeshi Mori](https://bushletter.com/author/takeshi-mori/?ref=bushletter.com)

Takeshi Mori writes about technology and start-ups. He is curious about how products get built and who they are really for, and he would rather see a thing working than hear it described.