Entity
Gemini 4 Argon
Gemini 4 Argon, Google DeepMind's latest frontier AI model, has rolled out to Gemini app Pro/Ultra subscribers and demonstrated autonomous multi-agent capability by defeating Minecraft's Ender Dragon without human intervention (arXiv, October 16). A separate community benchmark, StarSkirmish Bench, has also begun testing Gemini 4 Argon against flagship OpenAI, Anthropic, and Muse models on code-based StarCraft: Brood War play (StarSkirmish Bench, October 16).
This page was written and is updated by the paper from the source articles listed below. Importance is how central the topic is to current coverage; confidence is how well-sourced the page is; mentions counts the articles that fed it. How the wiki works →
## Gemini 4 Argon
**Type:** Product
**Status:** Rolling out in phases; now available to Gemini app Pro and Ultra subscribers (SiliconANGLE, October 13), after initially launching to vetted cybersecurity defenders roughly two weeks earlier (SiliconANGLE, October 13).
**Key Facts:**
- Google DeepMind announced Gemini 4 Argon as its latest frontier AI model (DeepMind, October 13).
- Google released "Gemini 4 Argon for Robotics," a variant specialized for robotics applications (Google Developers Blog, October 13).
- Initial access was limited to vetted cybersecurity defenders before wider release (SiliconANGLE, October 13).
- Gemini app Pro and Ultra subscribers now have access to Gemini 4 Argon (SiliconANGLE, October 13).
- The rollout includes a new "tasks mode" that allows Gemini to continue working on a job after the user steps away (SiliconANGLE, October 13).
- Researchers placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world with the goal of defeating the Ender Dragon; the agents coordinated via in-game chat and shared memory, with no assigned roles or pre-written game logic, and reportedly succeeded in beating the game (arXiv, October 16).
- Agents controlled players using the open-source Mineflayer library and autonomously planned, wrote, and debugged their own programs; human researchers could watch via a live dashboard and had the ability to pause or stop the run but reportedly found no cause to intervene (arXiv, October 16).
- A separate benchmark, StarSkirmish Bench, tested Gemini 4 Argon alongside flagship models from OpenAI, Anthropic, and Muse on playing StarCraft: Brood War by writing code; the project describes itself as a "Show HN" community benchmark refresh (StarSkirmish Bench / Hacker News, October 16). Specific performance results for Gemini 4 Argon were not detailed in the source.
**Timeline:**
- Approx. late September 2026: Gemini 4 Argon first made available to vetted cybersecurity defenders (SiliconANGLE, October 13).
- October 13, 2026: Gemini 4 Argon and the robotics variant formally announced (DeepMind; Google Developers Blog, October 13).
- October 13, 2026: Gemini 4 Argon rolls out to Gemini app Pro and Ultra subscribers with new tasks mode (SiliconANGLE, October 13).
- October 16, 2026: arXiv preprint describes a multi-agent Gemini 4 Argon team autonomously defeating the Ender Dragon in Minecraft, cited as a demonstration of emergent multi-agent coordination and autonomous coding capability (arXiv, October 16).
- October 16, 2026: StarSkirmish Bench publishes updated results comparing Gemini 4 Argon to flagship models from OpenAI, Anthropic, and Muse on a code-based StarCraft: Brood War benchmark (StarSkirmish Bench, October 16).
**Connections:** Gemini (Google), Google DeepMind, Physical Intelligence (robotics/embodied AI competitor), Gemini app subscription tiers (Pro, Ultra), Mineflayer (open-source Minecraft agent library), autonomous multi-agent AI research, StarSkirmish Bench (LLM gaming benchmark), OpenAI/Anthropic/Muse flagship models (benchmark comparison)
**Section Relevance:** ai-tech (high); business (moderate, re: subscription monetization); robotics (moderate); research (moderate-high, re: multi-agent autonomy and game-playing benchmarks)