Blog · AI Agent News · Agent basics VIEN
Gemini 4 Neon: Medium Costs 8× Less Than Claude, Scores Higher
Zapier ran AI through 657 simulated office tasks: Neon on Medium cost $0.58 a task, Claude Fable 5.2 cost $4.64 and scored lower. Which setting should you pick?

In this article
Say you have an agent that runs every night to sort your email and sync your spreadsheets (an agent is AI that does many steps for you, like an assistant who searches, compares prices, then books the ticket). I also tend to switch on the priciest model at its deepest thinking level, just to feel safe. Yet in Zapier's test, Claude Fable 5.2 cost about $4.64 a task and still scored below Gemini 4 Neon's $0.58 setting (51.5% vs 53.3%).
My take: pricier does not mean better. Start at Medium, and move up only when it gets things wrong. "Reasoning level" means how long the AI is allowed to think before it answers. The longer it thinks, the more it costs, like a phone plan billed per second.
One limit up front: this is a simulated test (like an AnTuTu score when comparing phones) that Zapier ran on November 17–18, 2026, not your real work. The prices are also estimates from list prices, not invoices.

Does paying 8× more buy a higher score?
No, at least not in this test. Zapier gave each setting 657 tasks in a simulated company and graded the final result. Neon on Medium got 350 tasks right, while Claude Fable 5.2 at its top setting got fewer right and cost 8× more.
| Choice | Score | Cost per task |
|---|---|---|
| Neon, Medium | 53.3% | $0.58 |
| Neon, XHigh | 56.3% | $1.69 |
| Claude Fable 5.2 (Max, with Anthropic's default fallback model) | 51.5% | $4.64 |
Going from Medium to XHigh costs nearly 3× more. What do you get?
Neon on XHigh scores 56.3%, the highest on the board. But measured from Medium, you pay nearly 3× more ($0.58 to $1.69) for about 20 extra tasks out of 657, roughly 3 points. At this level a failed task also costs more than a passed one, so raising the setting without measuring is an easy way to waste money.

In some places raising the setting is nearly pointless: on operations tasks, Neon's lowest setting already matches its highest. And HR tasks are Neon's weakest spot, where a higher setting does not rescue it.
Also readApple Music Eats 8–11 GB With Zero Downloads: How to Fix ItWhat to do about it
- Open your agent's settings and find which steps run at the highest level.
- Move the repetitive steps (sorting email, syncing spreadsheets) down to Medium, run them for a few nights, and compare against before.
- Raise only the steps that really keep failing to High or XHigh.
Google also says Neon on High costs roughly 50–80% less per task than other top models, but that is Google's own figure, measured on other leaderboards, not Zapier's.
The comfort of switching on the top setting is not cheap, and by this test it does not even buy a higher score. Next time before you switch it on, ask: does this step really need to think that long?
Sources
Agent basicsGoogle GeminiAutomation
Agent basics · Nov 22, 2026GPT-Live-1.1 Standby Cuts the Idle Rate 60%, Not Your BillStandby drops the idle-minute price from $0.05 to $0.02, but only after a timeout. We run 1,000 idle hours to show how…
Coding agents · Nov 22, 2026Moving to GPT-6.2 with your old prompt? Your agent may wait for "OK"OpenAI had to write three prompt snippets to stop GPT-6.2 from pausing to ask. Before an overnight agent runs on it, retest…
Safety & policy · Nov 21, 2026Nvidia Without Jensen Huang: Where Do CUDA and GPUs Go?Analyzing Nvidia after Jensen Huang: the 60-report leadership gap, Claude Mythos and AMD pressure on CUDA, and what GPU buyers face next.