Event · User reviews

Google's Gemini 4 Argon: what people who tried it said

Hands-on reviews posted on X in the first day. Posts from Google's official accounts and staff are left out, and both praise and criticism are included.

1. Positive@AiHubMix · X ↗

   Same prompt, 25 minutes vs 3

   API aggregator AIHubmix ran the same prompt against GPT-6 Astra: Astra took 25 minutes and cost $11, Argon took 3 minutes and cost $0.50. It said it was surprised by Argon's speed.

2. Positive@notjazii · X ↗

   Almost the same result as Opus 5.5

   One user ran the same prompt at the highest reasoning level and got nearly identical results. Opus took 15 minutes and cost $8.2 via Claude Code, Argon took 13 minutes and cost $1.2 via Devin CLI.

3. Positive@levie · X ↗

   Up 14 points on Box's eval

   Box CEO Aaron Levie said Argon jumped 14 percentage points over Gemini 3.1 Pro across all tests on Box's complex-work eval and is 'firmly at the frontier,' with +37 points in technology and +25 in healthcare.

4. Positive@Hesamation · X ↗

   No. 1 on a Stronghold Crusader test

   A user said Argon ranked first, well ahead of Opus 5.5, on his CrusaderBench, which measures how well models grow an economy in Stronghold Crusader. He said it pulled off tricks usually used by pros, such as building wells to grow the population.

5. Mixed@kieranklaassen · X ↗

   It's with the pack now

   Kieran Klaassen said Argon isn't better than the others but has joined the leading pack. Low effort is very good, he said, and it built a webcam app that reads heart rate off his forehead in 35 minutes. For design it's second behind Opus 5.5, but it doesn't check its own work as well.

6. Mixed@bindureddy · X ↗

   Above open source, below Astra and Fable

   Bindu Reddy said on her own benchmark Argon falls short of Fable, Astra and GPT 6.1 Sol, but is ahead of DeepSeek Flash and much better than Meta's Muse Wonder.

7. Negative@bridgemindai · X ↗

   Nowhere near Opus 5.5

   BridgeMind said it's a solid improvement for Google but comes nowhere near Claude Opus 5.5, and that it's disappointing Google is still this far behind.

8. Negative@danshipper · X ↗

   21% on a senior engineer test

   Every CEO Dan Shipper said Argon struggles on the company's senior engineer benchmark task, scoring 21%, far below Sol 6 and even Meta's Muse.

9. Negative@i2cjak · X ↗

   Useless for circuit boards

   One user said that on his PiNas test Argon flipped connectors and produced a nearly unreadable schematic, and told people to stick with Opus 5.5 or Sol 6.1 for electronics.