Gemini 3.5 Pro delayed a third time over hallucinations and inconsistent outputs

Gemini 3.5 Pro's third delay is becoming a story about competitive pressure as much as engineering. Reports say the model is failing to clear key internal benchmarks, with frequent hallucinations and inconsistent outputs, and remains locked in a limited Vertex AI enterprise preview — no model card, no public benchmarks, no official pricing. Google has touted a 2M-token context window and a 'Deep Think' mode, but neither is available for general access.

The new wrinkle is a rumored interim release: Gemini 3.6 Flash, a smaller/faster model that could satisfy immediate market demand while DeepMind keeps working on 3.5 Pro. That would be a tacit admission the flagship isn't ready as OpenAI's GPT-5.6 tier reaches full general availability and Anthropic's Fable 5 stays in market.

Context compounds the pressure. Google is simultaneously projecting strength elsewhere — Gemini Omni Flash debuted at #1 on Artificial Analysis's text-to-video and image-to-video leaderboards, and Google shipped Gemma 4 E2B for on-device Pixel use and faster Gemini Spark Workspace integrations. But a repeatedly delayed frontier text model undercuts the narrative in the segment developers watch most closely.

Competitively, the delay hands momentum to rivals mid-cycle: OpenAI is flooding channels (Bedrock, M365, GitHub) with GPT-5.6, and open-weight challengers are multiplying. Skeptics note that shipping a hallucination-prone flagship would be worse than delaying it, and that a Flash stopgap is a reasonable release-management call. What to watch: whether Google confirms 3.6 Flash, and whether 3.5 Pro ever ships with the promised Deep Think capability or gets folded into a later version.