7/25/2026, 1:02:33 PM · foundation-models

Google Ships Gemini 3.6 Flash, Confirms Gemini 3.5 Pro Still Delayed, and Discloses Gemini 4 Pretraining Has Begun

Google's July 21 model drop brings an efficiency-focused Flash upgrade and two companion models, while the flagship Pro variant remains in partner testing and the company publicly confirms it has started its most ambitious foundation-model training run to date.

What Shipped

On July 21, 2026, Google released three models simultaneously: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. <cite index="1-11">Google released Flash 3.6 alongside Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber on July 21, 2026.</cite>

<cite index="4-1">Gemini 3.6 Flash is optimized for multi-step orchestration, full-stack code refactoring, and general reasoning.</cite> According to Google's official developer documentation, <cite index="5-1,5-2">it provides sustained frontier-level intelligence at higher speed and lower cost, excelling at code generation, agentic execution, and spatial reasoning.</cite> The model retains the one-million-token context window of its predecessor while advancing its knowledge cutoff. <cite index="2-2">Its knowledge cutoff is March 2026, up from January 2025 on Gemini 3.5 Flash.</cite>

On benchmarks, <cite index="2-11">the model gains on coding — 58.7% versus 55.1% on SWE-Bench Pro — long-context retrieval, and computer use, reaching 83% versus 78.4% on OSWorld Verified.</cite> Critically for operators, <cite index="2-12">it uses about 17% fewer output tokens and drops the output price from $9.00 to $7.50 per one million tokens.</cite>

<cite index="3-8">Google also announced Gemini 3.5 Flash-Lite for high-throughput and low-latency tasks, like agentic search and document processing.</cite> The third addition, <cite index="3-10,3-11">Gemini 3.5 Flash Cyber, is designed for finding and fixing security vulnerabilities, with Flash serving as the foundation for detecting, validating, and patching code security issues at scale.</cite> <cite index="3-12">Access is initially for governments and trusted partners as part of a limited-access pilot program.</cite>

<cite index="3-1,3-2">Gemini 3.6 Flash and 3.5 Flash-Lite are available in the Gemini app, with developers able to access them through Google AI Studio and Android Studio.</cite> <cite index="2-8">Gemini 3.6 Flash also became available in GitHub Copilot on the same day as the announcement.</cite>

The Delayed Pro

Gemini 3.5 Pro — the higher-capability frontier Large Language Model (LLM) in the 3.5 series — was not part of the release. <cite index="11-1,11-2">In mid-May, Google announced Gemini 3.5 Flash at I/O 2026 and said the Pro version would arrive in June. That deadline has passed.</cite>

<cite index="11-5">According to Bloomberg, Google is "taking time to try to improve Gemini 3.5 Pro's capabilities, particularly in coding." In late June, Google updated the data being used to train Gemini in an attempt to improve coding skills, but the results were disappointing.</cite> <cite index="14-13">Ten current and former employees described frustration inside the company, with concern that Google is losing ground while Anthropic and OpenAI ship models that outperform Gemini.</cite>

In its official July 21 blog post, Google acknowledged the situation without offering a timeline. <cite index="7-14">The company stated that Gemini 3.5 Pro is currently testing with partners and that it plans to make it broadly available as soon as it is ready.</cite> The delay had market consequences: <cite index="15-2">the delay contributed to a roughly 4% drop in Google's share price.</cite>

Gemini 4 Pretraining Confirmed

The most consequential disclosure in the July 21 announcement was forward-looking. <cite index="7-15,7-16">Google stated that its team is already focusing on building the next generation of models and has started its most ambitious pre-training run yet for Gemini 4, expressing excitement about the progress.</cite>

Pre-training is the foundational, compute-intensive phase of building a foundation model. <cite index="21-9,21-10">During this stage, the model is trained on massive datasets to learn language, reasoning, coding, and multimodal capabilities before moving on to instruction tuning, reinforcement learning, and safety testing.</cite>

Alphabet Chief Executive Officer Sundar Pichai reinforced the disclosure on a subsequent earnings call. <cite index="24-3,24-4,24-5">Pichai stated: "We have started our most ambitious pretraining run yet for Gemini 4. I am very excited by the progress I'm seeing internally. I'm confident that people will be pleased."</cite> However, <cite index="18-4,18-5">Google has officially confirmed that Gemini 4 pre-training is underway and calls it the company's most ambitious pre-training run yet, but this is a development milestone, not a launch — Google has not announced a Gemini 4 release date, model sizes, benchmarks, pricing, context limits, API access, or a preview.</cite>

The simultaneous disclosure of a delayed Pro and an in-progress successor model drew industry attention. <cite index="20-12">By saying that Gemini 4's pre-training run has already begun, Google confirmed that development of its next flagship model is well underway even before Gemini 3.5 Pro has reached general availability.</cite>

Cross-references

Sources

  1. [1]
    Inside Gemini Flash 3.6, Google’s Long‑Context AI
  2. [2]
    Google Released Gemini 3.6 Flash - DataNorth AI
  3. [3]
    Google launches Gemini 3.6 Flash and 3.5 Flash-Lite, teases Gemini 4
  4. [4]
    Gemini 3.6 Flash | Gemini Enterprise Agent Platform | Google Cloud Documentation
  5. [5]
    Gemini 3.6 Flash | Gemini API | Google AI for Developers
  6. [6]
    ‎Gemini Apps’ release updates & improvements
  7. [7]
    Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
  8. [8]
    www.mexc.com
  9. [9]
    Google Reportedly Delays Gemini 3.5 Pro Over Coding Issues
  10. [10]
    Gemini 3.5 Pro Delayed to July 2026: What Developers Should Know | Bind AI
  11. [11]
    Gemini 3.5 Pro delays due to coding performance, upgraded Flash model in testing
  12. [12]
    Google Delays Gemini 3.5 Pro Launch as Coding Performance Doesn't Cut It
  13. [13]
    Alphabet's Gemini 3.5 Pro Is Late and the Stock Is Slipping. Is the AI Leader Falling Behind?
  14. [14]
    Google Delays Gemini 3.5 Pro Over Coding Issues: Report
  15. [15]
    Google's Gemini 3.5 Pro release faces delay: report | Seeking Alpha
  16. [16]
    Google Gemini 3.5 Pro faces "delays" over coding performance misses - Neowin
  17. [17]
    Gemini 3.6 Flash Cuts Token Costs and Scores Higher on Every Benchmark
  18. [18]
    Gemini 4 Training Has Begun: What Google Confirmed—and What It Did Not
  19. [19]
    Google starts pre-training Gemini 4, its most ambitious run yet
  20. [20]
    Gemini 3.5 Pro Is Testing With Partners, Have Begun Pre-training Run For Gemini 4 Pro: Google
  21. [21]
    Google Quietly Starts Training Gemini 4, Its Next Flagship AI Model
  22. [22]
    Chubby♨️ on X: "Google has begun pre-training Gemini 4, marking a completely new foundation model. This is really exciting! The announcement blog for 3.6 Flash states that Gemini 4 is being completely revamped. Presumably, the recent developments for 3.5 Pro were disappointing, so they're https://t.co/52GP9zQh5d" / X
  23. [23]
    Google launches Gemini 3.6 Flash, teases Gemini 4 pre-training | AI Weekly
  24. [24]
    Sundar Pichai Just Said Something About Google’s Gemini 4—and Sam Altman, Elon Musk Should Care
  25. [25]
    Google Starts Gemini 4 Pre-Training and Launches Efficient Agent Models / X