Google drops Gemini 4 Neon for coding
Google DeepMind released Gemini 4 Neon, a mid-tier model optimized for agentic software engineering and complex reasoning. Positioned between Hydrogen and Carbon, it combines near-frontier benchmark performance with a 1M-token context window at $3 per million input tokens.
Gemini 4 Neon targets the sweet spot developers actually care about: frontier reasoning quality at sub-flagship latency and cost.
- –Mid-tier efficiency: slots between Hydrogen and Carbon, cutting per-task costs by roughly 32% while matching Carbon on core coding evaluations.
- –Granular reasoning control: configurable thinking levels (low, medium, high, xhigh) allow developers to balance token spend dynamically between fast autocomplete and autonomous repo refactoring.
- –High steerability: ranks #2 on the Agent Arena Index for adapting cleanly to mid-task user steering, with Google's lowest internal fabrication rate yet.
- –Massive unified context: a 1M-token context window with 1M-token decode allows ingesting full codebases, logs, and multimodal assets in single prompts.
- –Aggressive pricing: priced at $3/1M input ($0.15 cached) and $15/1M output, directly undercutting competing frontier models across Google AI Studio, Antigravity, and Copilot.
DISCOVERED
3h ago
2026-11-20
PUBLISHED
3h ago
2026-11-20
RELEVANCE
AUTHOR
Theo - t3․gg