Gemini 4 Argon (gemini-4-argon) compared side by side with Gemini 3.8 Flash, Gemini 3.1 Pro Preview, Gemini 3.5 Flash-Lite, and Gemini 3 media specialists across context windows, 1M long decode, caching, pricing, breaking changes, and agent routing.
Lineup head-to-head
Every Gemini 3 model caps output at 64K. Gemini 4 Argon unifies input, thinking, and output in one 1M-token window with long decode up to 1M.
Argon vs 3.8 Flash
3.8 Flash ($0.75/$3.75 + Free tier) is the default Antigravity workhorse; escalate to Argon ($2/$10 intro) for deep reasoning or >64K outputs.
Argon vs 3.1 Pro
Argon replaces 3.1 Pro with flat pricing across 1M tokens—cheaper than 3.1 Pro ($2–$4 in / $12–$18 out) at every prompt size during intro pricing.
3.5 Flash-Lite & media
Use 3.5 Flash-Lite ($0.30/$2.50) for subagent fan-out, and pair Argon (text-out only) with 3.8-live, nano-banana-2.1, and omni-1.1-flash for audio/image/video.
Upgrade gotchas
Explicit caching (cached_content) throws an error on Argon, and temperature/top_p/top_k are ignored; implicit caching discounts >=4K prefixes by 95%.
3-tier routing
Route parsing to 3.5 Flash-Lite, daily coding loops to 3.8 Flash, and hard refactors, audits, or long-decode tasks to Gemini 4 Argon.