Google Is Back: Gemini 4 Argon Beats Opus 5.5 at Half the Price

Google DeepMind announced Gemini 4 Argon on September 30, its first new flagship in more than seven months. On Google's own chart it beats Claude Opus 5.5 on 13 of 17 tests, and GPT-6 Astra 13 times too, at half of Opus's price: $2 in and $10 out per million tokens. There are three catches. The chart is Google's own. You can't use Argon yet. And the half price is an introductory deal with no end date. On the independent Artificial Analysis Intelligence Index, Argon ties GPT-6 Astra at 53, five points behind Opus 5.5 at max. That's still a 12-point jump over Google's previous best, and a real comeback. Every number is shown on its source page, highlighted as it's spoken. What's covered: What shipped: the new naming, 1M output tokens (up from 64K), and Argon already at work inside Google Why Google needed this: the Gemini 3.5 Pro that never shipped, and Pichai's "early look" Google's 18-test chart: who picked the tests, and where the rival scores come from Where Argon wins: office work (AutomationBench), long documents (GraphWalks), legal, finance, biology, math, and video (with an asterisk) Coding: first on DeepSWE, last on FrontierSWE and Terminal-Bench 4.0 The test where Google gave Argon 6× the grading time, and it still came third The neutral scoreboard: Artificial Analysis, setting for setting The real price: launch vs after the intro deal, vs Opus, Astra and GPT-6.1 Sol Who can use it, and when Who should use it: long documents and office work vs terminal coding Would you switch from Claude or GPT to Argon at half the price, or wait for the independent tests? Tell me in the comments. Chapters 0:00 Argon beats Opus 5.5 on 13 of 17 tests 0:24 Three catches 0:35 Short version: Google's comeback 0:51 What we'll check 1:04 What Google actually released 1:42 Why Google needed this 2:09 Google's chart: who picked the tests 2:28 Office work 2:46 Long documents 3:16 Coding: DeepSWE 3:45 FrontierSWE and Terminal-Bench 4.0 4:09 6× the grading time, still third 4:30 The neutral scoreboard 5:08 Setting for setting 5:21 The comeback in one number 5:35 Is it really half the price? 5:58 The intro price, and when it ends 6:13 The fair rival on price: GPT-6.1 Sol 6:33 Can you use Argon? 7:03 Who should use it 7:24 My read 7:54 Your turn #Gemini #GoogleDeepMind #AI