Google Gemini 4 Argon Broadly Available, Leads Vals Index for Professional AI
Argon's broad availability gives enterprises and developers their first widespread opportunity to test Google's most capable publicly offered model against rivals from OpenAI and Anthropic. Its benchmark leadership in professional tasks combined with a significant price discount shifts the competitive calculus for organizations choosing AI platforms.
Gemini 4 Argon is broadly available to paying customers as of October 13, 2026, after a two-week restricted period limited to vetted cyber defenders and early-access partners following its September 30 unveiling. On October 16, Google DeepMind posted a demonstration in which four Argon agents operating in a shared Minecraft world wrote 2,783 programs, divided tasks via chat, shared notes, and debugged errors collaboratively, defeating the Ender Dragon in 27 hours with no human assistance.
On Vals AI's GDP-weighted index of professional agentic tasks, Argon leads at 68.90%, ahead of Claude Sonnet 5.5 at 67.04% and Claude Opus 5.5 at 66.97%. Its cost edge over Opus 5.5 stems from using fewer than half the input tokens, averaging $15.68 per test versus $32.14 for Opus 5.5, an advantage Google says should persist after introductory pricing ends. Argon leads DeepSWE v1.1 at 77.9%, Vibe Code Bench at 91.9%, and long-context GraphWalks at 84.2% versus Claude Opus 5.5's 66.8%. On science, it leads LABBench2 at 88.8% but trails GPT-6 Astra on Terminal-Bench Science 0.1, 57.6% to 68.1%. It leads Harvey's Legal Agent Benchmark at 19.6% versus GPT-6 Astra's 5.4%.
Coding is the consistent gap. Argon finishes last of four frontier models on FrontierSWE v2 at 55.0% and Terminal-Bench 4.0 at 57.4%, trailing Claude Opus 5.5 by more than eight points on the latter. On Code Arena WebDev, Argon ranks eighth with 1,679 points, trailing GPT-6.1 Sol's 1,759. One analysis characterized this as a deliberate strategic choice. On Artificial Analysis's independent Intelligence Index, Argon ties GPT-6 Astra at 53 and sits five points behind Claude Opus 5.5 at 58. Some Google employees reportedly believe Argon performs well on benchmarks but struggles with real coding and front-end design, which Google disputes.
Argon is priced at $2 per million input tokens and $10 per million output tokens through November 30, 2026, rising to $4/$20 after that date, with cached input tokens at 95% off. The model supports a 1 million token output context window, up from 64K on the prior generation. At introductory pricing, Argon is priced at roughly one-fifth of GPT-6 Astra's listed input price.
Google DeepMind posted a multi-agent demonstration on October 16 showing four Argon agents completing Minecraft's Ender Dragon challenge in 27 hours, writing 2,783 programs and coordinating entirely through text chat with no human help.
The model had reached broad availability for paying customers on October 13 after a two-week restricted period following its September 30 unveiling.
New usage and benchmark data published October 14 added nuance to Argon's performance following its October 13 broad rollout to paying customers.
AA-Omniscience found Argon hallucinates at 15% versus 51-54% for GPT-6 variants, but its higher abstention rate yields only 50% factual accuracy against GPT-6 Astra's 63%, and high output verbosity narrows the real-world cost advantage. A reported 11-hour unguided engineering run by Google engineers George Weale and Bo Yang on a 205,000-line codebase found Argon required explicit review rules to produce shippable output, with two of six proposed optimizations accepted. Some Google employees reportedly doubt Argon's real-world coding and front-end design quality, a characterization Google disputes.
Google opened Gemini 4 Argon to paying customers on October 13, two weeks after a September 30 unveiling restricted to vetted cyber defenders and early partners.
Benchmark analyses the same day show Argon first on the Vals AI index at 68.9% but last among four frontier models on FrontierSWE v2 at 55.0% and Terminal-Bench 4.0 at 57.4%, with Artificial Analysis scoring it tied with GPT-6 Astra rather than ahead. Introductory pricing of $2/$10 per million tokens is one-fifth of GPT-6 Astra's listed price.
Argon ranked eighth on Arena's Agent Arena leaderboard based on 3,417 sessions
Google unveiled Gemini 4 Argon with access restricted to vetted cyber defenders and early-access partners
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free