Gemini 4 Argon
Gemini 4 Argon is a closed AI model from Google with an index score of 191 (GPT-4 = 100), rank 8 of 217.
191176–207Index score#8 of 217 models
$2.0/ $10Price per million tokensinput / output · via OpenRouter
1.05MContext windowtokens it reads at once
30 Sep 2026Released2 months ago
Where it sitson the frontier
GPT-6 Astra reached this score 1 month earlier.
Test resultsby area · tick = best by any model
Reasoning & knowledgeevidence
Mathsevidence
Codingevidence
Agentsevidence
Professional workevidence
Novel problemsevidence
Long tasksno evidence yet
Third-party results only, at the model’s best standard setting. Hover a bar for who ran it.
What the lab says6 of 18 also run by third parties
| Test | Google says | Third parties |
|---|---|---|
| Vals Finance Agent v2 | 65.4% | 65.4%Vals AI |
| Harvey's Legal Agent Benchmark | 19.6% | 27.1%Vals AI |
| FrontierSWE v2 | 55.0% | 55.0%Proximal |
| Vibe Code Bench | 91.9% | 91.9%Vals AI |
| Terminal-Bench 4.0 | 57.4% | 57.6%Vals AI |
| Terminal-Bench Science 0.1 | 57.6% | 44.3%Vals AI |
| AutomationBench | 51.3% | not run yet |
| DeepSWE v1.1 | 77.9% | not run yet |
| PostTrainBench | 45.3% | not run yet |
| LABBench 2 | 88.8% | not run yet |
| RiemannBench | 76.0% | not run yet |
| GraphWalks, up to 128K tokensBFS, F1 | 99.7% | not run yet |
| GraphWalks, 256K–1M tokensBFS, F1 | 84.2% | not run yet |
| Agent's Last Exam | 39.5% | not run yet |
| OSWorld 2.0offline set, partial credit | 69.2% | not run yet |
| Chartography | 71.6% | not run yet |
| LVBench | 91.7% | not run yet |
| CWE-bench v1 | 68.0% | not run yet |
From Google’s launch post, Gemini 4 Argon, read from the chart image in the post. Shown, never scored; settings can differ from the third-party run.
What others sayrank on 10 leaderboards
The Model Indexthird-party tests, our method#8of 217 · 191
LMArena · Instruction followingfollowing detailed instructions#1of 362 · 1528LMArena · Creative writingcreative writing#1of 360 · 1523LMArena · Mathsmaths questions in chat#1of 346 · 1536LMArena · Long promptslong, complex requests#1of 340 · 1544LMArena · Textoverall chat quality, judged by people#2of 362 · 1522LMArena · Codingcoding questions in chat#2of 357 · 1553Epoch Capabilities Indexoverall capability from many benchmarks#10of 231 · 166.6LMArena · WebDevbuilding web apps and front ends#14of 133 · 1679LiveBench · Overalloverall, on regularly refreshed questions#10of 38 · 77.7Best attop three among current models
1stmaths questions in chatLMArena · Maths1stcreative writingLMArena · Creative writing1stfollowing detailed instructionsLMArena · Instruction following1stlong, complex requestsLMArena · Long prompts1streasoning questionsLiveBench · Reasoning1stFinance Agent (v2)Vals AI · third-party2ndcoding questions in chatLMArena · Coding2nddata analysis questionsLiveBench · Data Analysis2ndAPEX-AgentsMercor · third-party3rdARC-AGI-2ARC Prize · third-party
Where to use itprices checked 18 Nov 2026
API providers 2 · per 1M tokens, in / out
Host prices from OpenRouter; going direct can differ.
Facts and sources
Released30 Sep 2026
Announcementblog.google ↗
Release entryGemini 4 Argon
Scored asGemini 4 Argon
Output speed–
Hugging Faceclosed weights