All tasks / Comparisons
Gemini 4 Argon vs Gemini 3.8 Flash
Compared only on public benchmark results, across the 5 tasks where both are listed: Gemini 4 Argon is a tier higher in 2 of 5 shared tasks, they share a tier in 2.
Gemini 4 Argon
Google's flagship Gemini reasoning model with a 1M-token context window, labeled stable in the Gemini API.
- Vendor
- Weights
- Closed: API and apps only
- Released
- 13 Oct 2026
- API price
- $2 in / $10 out per 1M tokens
- License
- proprietary
- Last verified
- 18 Oct 2026
Gemini 3.8 Flash
Google's Flash tier of faster, lower-cost Gemini models, aimed at agents, coding and high-volume use.
- Vendor
- Weights
- Closed: API and apps only
- Released
- 2 Sep 2026
- API price
- $0.75 in / $3.75 out per 1M tokens
- License
- proprietary
- Last verified
- 18 Oct 2026
Task by task
| Task | Gemini 4 Argon | Gemini 3.8 Flash | Higher tier |
|---|---|---|---|
| Chat & assistants Models | S1,524 on LMArena Text Arena (overall, style control) · Cloud | S1,496 on LMArena Text Arena (overall, style control) · Cloud | Same tier |
| Writing Models | S1,516 on LMArena Text Arena: Creative Writing (style control) · Cloud | S1,487 on LMArena Text Arena: Creative Writing (style control) · Cloud | Same tier |
| Coding Models | B1,678 on LMArena WebDev / Code Arena (overall) · Cloud | C1,586 on LMArena WebDev / Code Arena (overall) · Cloud | Gemini 4 Argon |
| Browser & computer agents Models | –No public result yet · Cloud | –No public result yet · Cloud | — |
| Workflow automation Models | S77.5% on AutomationBench-AA (Artificial Analysis) · Cloud | B59.9% on AutomationBench-AA (Artificial Analysis) · Cloud | Gemini 4 Argon |
A tier spans a range of scores, and small gaps on a leaderboard are often within its margin of error, so only a different tier counts as ahead. How we rank.
Same benchmark, both results
| Benchmark | Gemini 4 Argon | Gemini 3.8 Flash | Snapshot |
|---|---|---|---|
| Artificial Analysis Intelligence Index Composite index over Artificial Analysis' own evaluation suite, higher is better | 53 | 41 | 18 Oct 2026 |
| AutomationBench-AA (Artificial Analysis) Score: average share of each task's objectives completed across 657 business-workflow tasks in 40 simulated SaaS apps via REST APIs; a task with any guardrail (business-rule) violation scores zero, higher is better | 77.5% | 59.9% | 19 Oct 2026 |
| LMArena Text Arena (overall, style control) Arena score (Bradley-Terry rating from pairwise human votes, style-controlled default view), higher is better | 1,524 | 1,496 | 15 Oct 2026 |
| LMArena Text Arena: Creative Writing (style control) Arena score (Bradley-Terry rating from pairwise human votes, creative-writing prompts only, style-controlled view), higher is better | 1,516 | 1,487 | 15 Oct 2026 |
| LMArena WebDev / Code Arena (overall) Arena score (Bradley-Terry rating from pairwise human votes on web-app generation), higher is better | 1,678 | 1,586 | 18 Oct 2026 |
| SWE-Bench Pro V2 (public set, full) Resolve rate on 642 validated long-horizon tasks from 11 copyleft repositories, re-graded on a pristine image under a locked offline protocol, higher is better | 99.2% | 94.9% | 19 Oct 2026 |