Gemini 4 Argon
Google's multimodal monster: sees, watches and listens better than anything, and writes up to 1M tokens back.
Overall Intelligence
#5
of 56
0.5 pts behind #1
12
races led
12
weeks at #1
88.2
score
- Released
- Oct 13, 2026
- 6 days ago
- Context
- 1M
- 1M max output
- Price
- $2 / $10
- per 1M in/out · $4.00 blended
- Speed
- 132 tok/s
- 5s to first token
- Modalities
- text · image · audio · video
- outputs text
- Weights
- Closed
- API and apps
The 30-second version
Google's multimodal monster. Gemini 4 Argon understands images, documents, video and audio better than anything else in the race, and it is a genuine frontier contender on reasoning and science too.
- 1Dominant lead on video and audio understanding; leads the multimodal race outright.
- 2Top of our graduate-science race and within a photo finish on overall intelligence.
- 3Very long outputs and a 1M-token context make it strong for document-heavy work.
- 4Priced at $2 / $10, the same as the mid-tier models it outperforms.
What you need to know
Purpose
A single model that can watch, listen, read and reason: the backbone for multimodal products and research assistants.
What Google optimised for
Native multimodality and long context, plus scientific reasoning, tightly integrated with Google Search, Workspace and Vertex AI.
Unique strengths
- Understands long videos and audio natively, not via transcripts.
- Excellent chart, diagram and document comprehension.
- Strong scientific reasoning and research skills.
Where it ranks
Current snapshot · illustrative scores · click any race to watch it
Gemini 4 ArgonBest on each axis
Overall
Coding
Hard Reasoning
Science
Writing & Language
Multimodal
Tool Use & Agents
Speed & Cost
Overall Intelligence over time
Best use cases
01
Video and meeting analysis
Summarising, searching and extracting moments from hours of footage or calls.
02
Document intelligence
Reading scanned PDFs, charts, forms and slide decks accurately.
03
Scientific research help
Literature analysis, hypothesis generation and data interpretation.
Strengths
Video understanding
Dominant lead in our video-understanding race.
Science
Leads the graduate-level science race.
Documents
Top of the document-understanding race.
Weaknesses
Agentic coding
A few points behind the Claude models on long autonomous coding.
Verbosity
Can be wordy unless you set length limits.
Prompts that play to its strengths
Meeting video digest
Watch this 90-minute product review recording. Give me: (1) the five decisions made, with timestamps, (2) open questions nobody answered, (3) moments where someone seemed to disagree but did not say so explicitly, with timestamps and what you noticed.
Why it works: Uses native video and audio understanding, including tone and body language.
Chart-heavy report
Here is our 60-page quarterly report as a PDF. Extract every chart's underlying numbers into tables, flag any chart whose visual contradicts the text, and summarise the three most important trends.
Why it works: Plays to document and chart comprehension.
When to choose it, and when not to
Inputs include video, audio, scans or complex charts.
The work is multimodal or science-heavy rather than long agentic coding.
Pick something else when…
- Pure agentic coding marathons: Claude models are steadier.
- When you need the most natural prose: GPT-6.1 Sol.
Update history
The distilled version first; open any release for the deeper breakdown.
The Gemini Pro family