Three frontier models, three shared benchmarks: Gemini 4 Argon, GPT-6.1 Sol and Claude Opus 5.5 side by side
With Gemini 4 Argon generally available since October 13, all three frontier labs finally have a current flagship on a paid API key. A new Digital Applied comparison finds only three benchmarks carry vendor-reported scores for all three models, and none of them produces a clean winner.
What happened
Digital Applied published a three-way comparison of Gemini 4 Argon, OpenAI's GPT-6.1 Sol and Anthropic's Claude Opus 5.5 on October 15, drawing prices, context windows, output caps and benchmark scores from each lab's own documentation fetched that day. The key finding: for the first time this autumn, all three labs sell a current flagship model on an ordinary paid API key, and exactly three benchmarks have a vendor-reported score for all three models. Terminal-Bench Science 0.1: Argon 57.6 percent on Google's self-computed run, GPT-6.1 Sol 57.0 percent at max effort, Opus 5.5 at 58.7 percent from Anthropic versus 63.3 percent on the public leaderboard that Google and OpenAI relay. AutomationBench: Argon 51.3 percent, Sol 36.1 percent at max effort, Opus 5.5 40.0 percent self-reported versus 42.5 percent with fallbacks. DeepSWE v1.1: Argon 77.9 percent, Sol 75.2 percent at high effort, Opus 5.5 74.2 percent as a five-trial average. Prices are equal on paper: Argon and Sol list $2 input and $10 output per million tokens, Opus 5.5 lists $4 and $20. The footnotes change the math: Argon's cheap rate expires November 30 and doubles to $4 and $20 on December 1, Sol's $2 and $10 rate applies only to prompts up to 272K tokens, and Opus 5.5 keeps one rate across its whole 1M window.
Why it matters
The comparison refuses to name a winner, and that is the honest point. When scores are this close, the lab that ran the benchmark, the effort level, and whether fallback models were allowed move the numbers more than any real capability gap. The practical decision levers are elsewhere: price expiry dates, prompt-size thresholds, output caps per request (Sol and Opus 5.5 return up to 128K tokens per request while Argon caps non-streaming requests at 32,768), knowledge cutoffs, and whether your pipeline feeds video or audio directly to the model, where Argon is the only candidate of the three. Verified October 15, 2026.
Sources
Get updates like this every morning
- ② Card on Stripe
- ③ 7 days free
Then $2/month · cancel anytime in one click