Gemini 4 Argon and Claude Opus 5.5 don’t have a single winner across these evaluations. Journal du Net favored Opus on French prose and motion design, while Artificial Analysis scored Opus higher on its Intelligence Index but Argon lower on its weighted cost per task. A separate one-shot kart-game run by AnassKartit produced a different, task-specific result.
What the practical tests found
In its October 15, 2026 comparison, Journal du Net evaluated a landing page, a 500-word French story, and a 10-second promotional video made with Remotion. It judged Claude Opus 5.5’s French prose the most natural and its motion-design video the strongest. Gemini 4 Argon came close on the video, but the outlet found it visually overloaded. GPT-6 Astra produced the landing page the outlet considered most complete.
Across those three runs, Journal du Net reported standard-rate totals of $5.52 for Gemini 4 Argon, $1.43 for Claude Opus 5.5, and $2.50 for GPT-6 Astra. The individual task bills show where the gap came from: Argon cost $3.76 for the landing page and $1.71 for the video, compared with $1.07 and $0.30 for Opus. Their French-writing runs cost $0.05 and $0.06, respectively. Those totals belong to these specific prompts and runs.
How Artificial Analysis scores the two models
Artificial Analysis’s comparison uses Gemini 4 Argon High and Claude Opus 5.5 Max, Default Fallback. It gives Opus the higher Intelligence Index, while Argon has the lower weighted cost per task.
| Model and configuration | Intelligence Index | Weighted cost per task (USD) |
| Gemini 4 Argon High | 53 | $1.99 |
| Claude Opus 5.5 Max, Default Fallback | 58 | $5.98 |
The same comparison reports 78% for Argon and 70% for Opus on AutomationBench-AA. So the result changes with the metric: Opus scores higher on the Intelligence Index, while Argon scores higher on AutomationBench-AA and has the lower weighted task cost. NeoTeo’s earlier Intelligence Index coverage offers context for that composite score.
These cost figures describe different things. Journal du Net adds the bills from three specified task runs; Artificial Analysis reports a weighted cost per task for its evaluation. Neither is a universal price for getting a particular job done.
Google’s September 30, 2026 announcement set Gemini 4 Argon’s introductory API rates at $2 per million input tokens and $10 per million output tokens through November 30. Google said standard rates of $4 per million input tokens and $20 per million output tokens would apply starting December 1. The announcement also set a 95% discount on cached input tokens. These are token rates, not the total cost of a finished task. Google’s announcement gives the pricing terms.
What a one-shot kart-game run adds
In an October 13, 2026 video, creator AnassKartit reported that Gemini 4 Argon completed a one-shot 3D kart-game task in 221 seconds for $0.31. GPT-6 Astra took 323 seconds and cost $0.70; Claude Opus 5.5 hit a 32,000-token limit and was cut off. This is one creator-run task, not a general model ranking.
Which results fit your work?
For French prose or a Remotion motion-design video, Journal du Net’s judgments are the closest match among these examples. Artificial Analysis’s scores answer a different question, using its named model configurations and evaluation metrics. AnassKartit’s result is specific to a one-shot game build and its 32,000-token cutoff for Opus.
When you compare models for your own workflow, start with the task you actually need done. Run the same deliverable in the tools and configurations you plan to use, then judge the output and the full task bill together.
