What's going on with Gemini?

2026-05-29 · Source: Martin Alderson · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & & Engineering · Depth: Intermediate, medium

Summary

Google's AI strategy, particularly concerning Gemini, appears distinct from competitors like Anthropic and OpenAI, despite Google's deep research, custom silicon, and vast resources. While Anthropic and OpenAI lead in frontier model intelligence with models like GPT5.5 and Opus 4.8, Google's Gemini 3.1 Pro lags slightly in benchmarks, and even behind some Chinese models (GLM 5.1, Qwen 3.7) for software engineering tasks. The recent Gemini 3.5 Flash, announced at Google I/O, exhibits underwhelming coding benchmarks but is exceptionally fast, achieving 206 tokens/second, roughly 4x faster than Opus 4.8 and GPT-5.5. However, its price increased significantly to \$9/MTok, making its external market fit unclear. This model likely targets Google's immense internal token consumption, leveraging its custom TPU 8i hardware for inference efficiency. A key weakness for Google is its fragmented coding agent strategy, with tools like Antigravity, Jules, Gemini Code Assist, and AI Studio, which contrasts with unified offerings like Claude Code and Codex, potentially hindering telemetry and training data collection.

Key takeaway

For AI Engineers evaluating large language models for user-facing applications, you should consider Gemini 3.5 Flash's exceptional speed (206 tokens/second) despite its higher cost (\$9/MTok) and mid-pack coding benchmarks. Its design suggests Google optimizes for internal, high-volume use cases leveraging custom hardware. However, if your team relies on integrated coding agents, Google's fragmented tooling ecosystem (Antigravity, AI Studio) presents a significant weakness compared to unified offerings like Claude Code or Codex, potentially impacting your development workflow and data collection.

Key insights

Google's Gemini strategy prioritizes internal token consumption and hardware integration over direct competition in external frontier model intelligence.

Principles

• Hardware-software co-design yields efficiency.
• Internal consumption drives model design.
• Unified developer tools gather data.

In practice

• Evaluate models on speed for user-facing apps.
• Consider self-hosting Chinese models for cost.
• Prioritize integrated developer tooling.

Topics

• Gemini 3.5 Flash
• Large Language Models
• AI Hardware
• Inference Efficiency
• Coding Agents
• Google AI Strategy

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Director of AI/ML, AI Scientist

Related on AIssential

• Gemini 3.5 Flash: Google's Most Powerful Model Ever! Beats Opus 4.7 & GPT 5.5? (Fully Tested) — Gemini 3.5 Flash offers superior coding and creative generation at a higher cost and token usage than previous flash models.
• Sundar Pichai admits Google trails in AI coding — Google acknowledges trailing in AI coding, stressing larger base models and improved agentic capabilities are vital for future competitiveness.
• How Google’s New Gemini Flash Models Compare to its Rivals — Google's new Gemini Flash models deliver competitive speed, cost-efficiency, and specialized capabilities for high-volume AI agent deployment.
• Gemini 3.5: frontier intelligence with action — Gemini 3.5 Flash delivers frontier agentic and coding performance at high speed and reduced cost.
• TAI #206: Gemini 3.5 Flash Is Stronger, But Flash Is No Longer Cheap — Google's latest Gemini models offer strong capabilities but come with significantly increased costs and deployment challenges.
• Gemini 3.5 Flash Looks Good For How Fast It Is — Gemini 3.5 Flash offers high speed for agentic tasks, but its intelligence and cost-effectiveness are debated.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Martin Alderson.