Question about the gemini future

Gemini API
1

Subject: Plans for extended test-time compute and long-horizon autonomous reasoning in Gemini?

Hi Gemini & DeepMind team,

Currently, when testing Gemini models on complex, multi-step engineering and logic tasks, the model appears heavily optimized for high-throughput and low latency (often spending just a few seconds before streaming output).

While this speed is great for general chats and context ingestion, it struggles with deep autonomous execution where competitor models allocate significant test-time compute (extended thinking loops, autonomous self-debugging, and multi-step verification before returning an answer).

Could you share any updates or roadmap details regarding:

  1. Configurable test-time compute (e.g., allowing users/agents to set a dedicated thinking budget of several minutes for complex logic/code verification)?
  2. The rollout of long-horizon agentic environments (like Jules / Project Astra features) to consumer or standard developer tiers?
  3. Whether Gemini 4 is primarily architectural-scale or if deeper test-time reasoning is being prioritized?

Thanks!