Gemini 4 Argon read 20 excerpts of my podcast transcripts for $0.28. Claude Opus 5.5 charged $0.91 and GPT-6 Astra $1.12 for the same job. Same prompt, default settings. On this job, Argon beat Opus on price across all 20 requests. On Dec 1, Argon's price doubles. Each of the three blind judges, one per vendor, preferred its own vendor's model. All four models scored between 4.51 and 4.70 out of 5. No human graded them. I didn't test code. Theo's codebases are a different job.