Claude or Gemini? The gap is smaller than the numbers suggest

Which is better, Anthropic's Claude or Google's Gemini? The answer depends on which models you compare and what you need them for.

BenchLM, a service that collects test results and prices, gives Claude Opus 5.5 an overall score of 85.4 and ranks it number one. Gemini 4 Argon gets 80.6 and sits at number six. The gap is therefore about 5 points. BenchLM has not disclosed how it calculates the overall score, and the Argon figure is only marked "estimated." That's because the model is available only in limited preview.

Compare the two models in the same price class, and the gap is smaller: Claude Sonnet 5.5 gets 82 at BenchLM, while Gemini 4 Argon gets 81. Gemini 3.8 Flash gets 72.

Coding: The numbers come from the developers themselves

On the Terminal-Bench 4.0 coding test, Claude Sonnet 5.5 scored 70.6 percent, Claude Opus 5.5 scored 66.4 percent, Gemini 4 Argon scored 57.4 percent and Gemini 3.8 Flash scored 19.1 percent.

However, these are results reported by the developers themselves, as relayed by the recruiting firm Second Talent. They are not independent tests, so the figures should be read with that in mind.

Price: Only Flash is clearly cheapest

Prices are per million tokens, the small chunks of text that AI models work in. The first figure is for what you send in, the second for what the model writes.

- Gemini 3.8 Flash: $0.75 / $3.75 - Claude Sonnet 5.5: $2 / $10 - Gemini 4 Argon: $2 / $10 - Claude Opus 5.5: $4 / $20

Gemini 3.8 Flash is therefore the cheapest. Gemini 4 Argon costs the same as Sonnet 5.5, and $2/$10 is an introductory price. It later rises to $4/$20, which is Opus level. Argon is thus cheaper than Opus only at the introductory price. The Flash price is scheduled to double to $1.50 / $7.50 on January 1, 2027.

According to Artificial Analysis, as relayed by Second Talent, an indexed task costs $5.98 with Opus 5.5, $1.99 with Gemini 4 Argon (introductory price) and $1.24 with Gemini 3.8 Flash. Flash is also faster: about 242 tokens per second versus about 97 for Opus.

What do the sources recommend?

BenchLM concludes that Claude dominates the rankings for coding and agent tasks, where AI works independently through multiple steps. The service recommends Opus 5.5 when a mistake costs more than the model, and Gemini where cost and capacity matter most. Flash models should be tested before they are used in autonomous workflows.

Second Talent similarly concludes that Claude is still best for complex tasks with long-term planning, while Gemini Flash is best suited to large volumes of well-defined tasks.

For ordinary users, the takeaway is that a single number rarely decides the choice. Price, speed and task type matter at least as much.