Google delayed Gemini 3.5 Pro in July 2026 after internal testing showed the model fell short of expectations in coding performance and long-horizon reasoning. CEO Sundar Pichai had announced the model would launch in June at Google I/O in May. It didn’t.

The delay signals trouble inside Google’s AI division. Engineers, researchers, and managers are frustrated. The model’s coding abilities lag behind systems from OpenAI and Anthropic. Internal conflicts and researcher departures have slowed progress.

What Went Wrong

Google updated Gemini 3.5 Pro’s training data late last month to improve coding performance. The results were disappointing. The model still failed to match competitors on key benchmarks that matter to developers.

Coding is the most commercially valuable use case for large language models right now. Finance, healthcare, and enterprise software all rely on AI-assisted development. If Gemini can’t compete here, it loses deals.

The Competitive Pressure

OpenAI released GPT-5.6 Sol in recent months. Anthropic released Claude Fable 5. Both improved on coding tasks. Moonshot released Kimi K3, which topped some coding benchmarks. Google’s inability to keep pace is becoming visible in benchmark results and in business wins.

Alphabet’s stock fell about 4% on the news of the delay. Investors see a company struggling to maintain leadership in AI, despite inventing the transformer architecture that underpins modern LLMs.

What Google Says Next

Google released Gemini 3.6 Flash on July 21 as a stopgap. Flash is faster and cheaper than Pro, but not more capable. It’s a holding action while Google continues to debug Gemini 3.5 Pro.

Google built the AI field. Now it’s playing catch-up. That’s a stunning reversal, and it’s visible in internal morale and external performance.