Google releases Gemini 4 Argon model; four agents autonomously beat Minecraft

1. #0109.30
   Initial2026-09-30 23:43 · publication time

   Google releases Gemini 4 Argon, called its most powerful model yet

   Google released Gemini 4 'Argon', which it markets as its most powerful model to date and pitches as a workhorse for coding and cybersecurity tasks.

   Source evidence: TechCrunch AI · QbitAI 量子位 · The Verge AI

2. #0210.01
   Confirmation2026-10-01 00:22 · publication time

   Google重回第一梯队：Gemini 4 Argon追平GPT-6 Astra

   Artificial Analysis' independent evaluation finds Google's Gemini 4 Argon ties GPT-6 Astra on the Intelligence Index at 53 points with a far lower hallucination rate (15%) and cheaper introductory API costs, though full-price costs and answer accuracy lag Astra.

   Source evidence: theblockbeats

3. #0310.13
   Confirmation2026-10-13 16:03 · publication time

   Gemini 4 Argon

   Google DeepMind released Gemini 4 'Argon', a new frontier model available via API at introductory pricing of $2 per million input tokens and $10 per million output tokens (95% off cached input), drawing attention for its SVG generation and natural-language writing quality.

   Source evidence: 737max

4. #0410.14
   Escalation2026-10-14 00:37 · publication time

   谷歌Gemini 4 Argon正式开放：看段视频就能教机器人投沙包，连中7次

   Gemini 4 Argon expanded from limited availability for selected cybersecurity partners to a general opening for all global paid users; it is available in the Gemini App (Google AI Pro/Ultra), Antigravity (Ultra only), and paid API. Introductory API pricing of $2/$10 per million tokens lasts until Nov 30, after which it doubles. Google also showcased video-driven robot skill learning, simulation generation, and an 87% ERQA score.

   State after update: Google DeepMind's Gemini 4 Argon is now open to global paid users after initially being limited to selected cybersecurity partners. It is available via the Gemini App (Google AI Pro/Ultra), Antigravity (Ultra only), and paid API; introductory API pricing is $2 per million input tokens and $10 per million output tokens until Nov 30, after which it doubles. Google also showcased video-driven robot skill learning, automated simulation generation, and an 87% score on the ERQA physical-reasoning benchmark.

   Source evidence: theblockbeats

5. #0510.17
   Confirmation2026-10-17 05:33 · publication time

   谷歌4个Gemini 4 Argon通关《我的世界》：27小时死605次，花费约520美元

   Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools.

   Source evidence: theblockbeats

All events · Back to the feed