- Third miss. Google DeepMind failed to deliver Gemini 3.5 Pro by its July 17 target — the third consecutive deadline the flagship model has missed since CEO Sundar Pichai promised a June launch at Google I/O in May 2026.
- Coding problem. Bloomberg, citing ten current and former Google employees, reported the model is “months behind internal targets” because a late-June training-data update aimed at improving coding performance “fell short of expectations.”
- Market hit. Alphabet’s share price fell 4.44% on July 16 to close at $354.46, erasing approximately $200 billion in market capitalisation in a single session, though shares recovered partially to $357.51 in after-hours trading.
The latest delay to Google DeepMind’s flagship AI model follows a pattern that has tested investor patience since the start of the year. At Google I/O in May, Sundar Pichai told developers to “give us until next month,” framing June as Gemini 3.5 Pro’s arrival date. June passed without a launch. A July 17 target circulated through tech-news channels, but a Bloomberg investigation published on July 16 confirmed the model remains unready.
What broke
The specific problem, according to Bloomberg’s reporting from ten current and former Google employees, lies in coding performance. In late June, Google updated the model’s training data in an attempt to sharpen its coding capabilities. The results disappointed: the updated model did not meet Google’s own internal benchmarks, and no new training run has yet produced an acceptable result. An Alphabet spokesperson confirmed that the company is “currently testing 3.5 Pro, an upgraded Flash model, and other models with partners,” but offered no revised launch date.
The model’s earlier structural failure modes — difficulties maintaining consistency in complex SVG layouts and breakdowns in recursive tool-calling environments — were what prompted Google DeepMind to scrap the original build and restart. That decision, previously covered here when the July 17 date was first set, has evidently not resolved every weakness the rebuilt architecture introduced.
Competitive pressure intensifies
The timing of the Bloomberg report compounded the damage. OpenAI released GPT-5.6 Sol on July 8–9, emphasising coding capabilities. Meta released Muse Spark 1.1 on July 9 with similar positioning. Both announcements arrived in the days immediately before the delay news, sharpening the contrast between Google’s execution difficulties and its rivals’ shipping pace. Prediction markets on Polymarket assigned an 81% probability to a July 31 delivery date as of July 16, with August 7 also emerging as a leading alternative at 73%.
Gemini 3.1 Pro, released in February 2026, remains the current production model available to developers. Google has registered model names including Gemini 3.6 Flash and Gemini 3.5 Flash Light, suggesting an intermediate release may be used to bridge the gap before the Pro version clears its internal bar.