Gemini 4 Argon added to BenchGecko · 1,048,576 context, $2/$10 per M introductory
Google made Gemini 4 Argon generally available on 2026-10-13. BenchGecko lists it at $2/$10 per M tokens (introductory) with a 1,048,576 token context window.
What landed
Gemini 4 Argon is Google's flagship reasoning model. Google DeepMind announced it on 2026-09-30 for trusted testers and made it generally available on 2026-10-13 through the Gemini API paid tier, AI Studio, Gemini for business, Google Antigravity and the Gemini app. BenchGecko lists the row as gemini-4-argon, alongside a separate batch row.
Specs
From Google's model card:
- Inputs · text, image, video, audio, PDF
- Output · text
- Context window · 1,048,576 tokens, a unified input and output limit
- Max output · 65,536 tokens by default, up to 1,048,576 with long decode
- Thinking levels · low, medium, high
- Tools · function calling, structured outputs, code execution, Search and Maps grounding, URL context, file search, computer use (preview)
- Caching · implicit only
- Consumption · Batch, Flex and Priority supported
Pricing
Google lists an introductory price through 2026-11-30:
- Input · $2 per M tokens
- Output · $10 per M tokens
- Cached input · 95% off input, $0.10 per M tokens on BenchGecko
After 2026-11-30 the price moves to $4 input and $20 output per M tokens. BenchGecko currently stores the introductory $2/$10 rate with a 1,048,576 context window.
Vendor-reported results
Figures below are from Google's 2026-10-13 post. They are vendor-reported, not BenchGecko tests.
- Vals Index · ranked first (no score given in the post)
- ERQA embodied reasoning · 87%
- Gray Swan Indirect Prompt Injection · 0.7% attack success across 15 adaptive attempts
- Gamow Labs private evals · RareBench 57%, LabBench 52.2%
Google also reports internal engineering use: agentic code submissions up more than 35% in Antigravity, a 2.85x end-to-end inference speedup on Gemma via vLLM, and Agent Development Kit changes that cut in-memory session prep time by 40% and sped up cold imports by 21%.
Sources
- Gemini 4 Argon announcement
- Gemini 4 Argon model card
- BenchGecko model page: /model/gemini-4-argon
Source: BenchGecko
Written automatically from BenchGecko's tracked data for this period; every number comes from the changelog, model, mindshare and pricing datasets. Checked against the source data before publishing.