Gemini 4 results
Gemini 4 Argon, Carbon and Hydrogen on our coding briefs
Google made Gemini 4 Argon generally available on October 13, 2026 and Gemini 4 Carbon on November 10, 2026. We run each model through the same published one-shot briefs every model in the arena gets, once per task, and post every complete output as it was generated. Each card opens the real page, not a screenshot.
- Flex price per 1M tokens
- $1 / $5; Hydrogen $0.75 / $3.75
- Input / output per 1M tokens. Argon and Carbon introductory price until Nov 30, 2026, then $2 / $10.
- Thinking level
- medium
- Set on every request; thinking is billed as output
- Outputs live
- 13
- Across 9 challenges
One landing page per model
Same brief (Driftwood, a travel photo book SaaS page), same request settings. Cost is the measured Flex charge and tokens include thinking where the API reports it.
Same tasks, both models
Only challenges where both Argon and Carbon finished with a complete page.
| Challenge | Argon tokens | Carbon tokens | Argon cost | Carbon cost | Compare |
|---|---|---|---|---|---|
| 1 · Landing | 26.6k | 35.1k | $0.133 | $0.176 | Open |
| 2 · Game | 21.5k | 24.4k | $0.108 | $0.122 | Open |
| 4 · Dashboard | 22.3k | 34.9k | $0.112 | $0.175 | Open |
| Total | 70.4k | 94.4k | $0.353 | $0.473 |
Tokens are billed output tokens. Carbon reports thinking tokens separately and they are included; the API did not report thinking tokens for Argon, so its figure is the billed output. One run per model and task (n=1), so read differences as observations, not rankings.
Every Gemini 4 output so far
Filter by model or challenge. New runs show up here after they are checked and published.
1 · Landing
2 · Game
3 · Tool
4 · Dashboard
5 · Particles
6 · 3D Game
7 · CHIP-8
8 · News
12 · Digger
How these were made
Each challenge runs once per model through the Gemini API on the Flex tier with thinkingLevel MEDIUM and maxOutputTokens 65,536, which includes thinking. The published briefs are shared; execution routes use different wrappers. Recorded instructions appear in the arena details where audited. No tools, no retries for a better result and no edits to the output.
Cost is the charge computed from the usage the API returned at the Flex price in force on the run date. We do not show generation time: Flex requests can wait in a queue, so wall-clock time would mix waiting with generation.
A run that fails, is cut off or ends without a complete page is recorded, not rerolled, and does not appear in this gallery.
/method