Two of the most expensive AI models in the industry were handed the same assignment: build the Eiffel Tower. Claude Opus 5.5, Anthropic's flagship, spent two hours on the task and ran up a $33.45 bill. Gemini 4 Argon, Google's newly released model, finished in a single pass for $0.23 — approximately 145 times cheaper.

The Eiffel Tower showdown, by the numbers

The size of the gap cannot be explained by sticker prices alone. Gemini 4 Argon launched with introductory pricing of $2 per million input tokens and $10 per million output tokens, a rate that runs until November 30, 2026.

Claude Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens. Anthropic's pricing represents a 20% decrease from its previous rates. On a per-token basis, Argon is half the price of Opus 5.5.

Half price, however, does not add up to a 145-fold difference. The rest of the gap comes from how much work each model did to reach the finish line. A model that nails a task on the first pass burns far fewer tokens than one that iterates for two hours — and tokens are what buyers ultimately pay for.

Beyond the tower: benchmarks and specs

Google DeepMind unveiled Gemini 4 Argon on September 30, 2026, and the model became generally available on October 13, 2026. Anthropic shipped Claude Opus 5.5 days earlier, on September 22, 2026.

In Google's report, Argon led with a 68.9% score on the Vals Index. Across benchmarks shared between the two models, Argon came out ahead in 14 of 18 categories, with its strongest showings in legal and finance applications. It also posted solid results on agentic coding tests.

Opus 5.5 scored higher on independent aggregates such as the Artificial Analysis Intelligence Index. It also performed well on terminal-heavy benchmarks, which measure a model's ability to operate inside command-line environments.

That split matters when weighing the claims: Argon's headline score comes from Google's own report, while Opus 5.5's edge shows up on independent aggregates — meaning each model's strongest numbers come from a different kind of source.

The two models also differ in capacity and design. Argon supports up to 1 million tokens for unified input and output, meaning it can produce much longer finished work in a single response — a feature that matters for jobs like drafting lengthy documents or generating large codebases. Opus 5.5 caps output at 128,000 tokens and instead emphasizes adaptive reasoning, adjusting how hard it thinks based on the problem in front of it.

Why a $0.23 tower matters

The Eiffel Tower result reframes the competition around cost per task rather than cost per token. A model with a lower per-token price can still be the more expensive option if it needs many more passes to finish the job.

Opus 5.5 has already cut its prices by 20%, but if a rival can complete certain jobs on the first try, list-price reductions may not close the gap on real workloads.

There is also a timing caveat. Argon's $2 and $10 rates are introductory and expire on November 30, 2026. Enterprises budgeting around those figures should plan for what happens once the promotion ends, since today's comparison is measured against promotional pricing.

Source: CryptoBriefing