Gemini 4 Argon
Google DeepMind’s October 2026 flagship — 1M native multimodal context and $2/$10 introductory pricing, with a published Round 6 Starlight Arena receipt.
Read the full Gemini 4 Argon analysisContext
1.05M
Max output
66K
Input /1M
$2.00
Output /1M
$10.00
Live pricing via OpenRouter
Best for
- Whole-repository + video/image/PDF synthesis in a 1,048,576-token window
- High-throughput TypeScript/Next.js and UI component builds at $2/$10 ($0.10 cached input)
- Multi-agent tool routing with adjustable thinking depth (low, medium, high)
Watch out
Introductory $2/$10 ($0.10 cached) pricing runs through 30 November 2026. In Starlight Arena Round 6, Argon required thinking=high to catch subtle governance/provenance pushback prompts — at thinking=low it optimizes for speed over specification critique.
For creators. The new default for 1M-context multimodal ingestion, video-to-code UI synthesis, and warm-cache repository sessions; keep Claude Opus 5 or Fable 5.1 on long-form literary prose and adversarial review.
Capabilities
- Text, image, audio, video and PDF input; text output with adjustable thinking levels (low, medium, high)
- 1,048,576 input-token window; 65,536 default output limit expandable to 1,000,000 tokens via long decode continuation
- 95% context-caching discount ($0.10 per 1M cached input tokens) for warm repository and document sessions
- Function calling, structured outputs, code execution, search grounding and computer use preview
- Measured in Starlight Model Arena Round 6 (/research/arena-receipts/2026-10-14-r6-gemini-4-argon-proving-ground.json)
Compare Gemini 4 Argon
Claude Opus 5 vs Gemini 4 Argon
Both passed 5/5 tasks in Starlight Arena Round 6. Route whole-repo + video/PDF multimodal workloads and warm-cache loops to Gemini 4 Argon at 40% of the token cost; route adversarial review, ambiguous specs, and long-form editorial voice to Claude Opus 5.
Gemini 4 Argon vs GPT-5.5
Gemini 4 Argon passed 5/5 tasks in Starlight Arena Round 6 against GPT-5.5’s 3/5 (2 partials on WCAG contrast and strict word-window cap), while costing 60–67% less per token. Keep GPT-5.5 where Codex-native terminal or OSWorld desktop loops are already qualified.
Claude Sonnet 5 vs Gemini 4 Argon
Price-matched at $2/$10 per million tokens and tied at 5/5 in Starlight Arena Round 6. Pick Claude Sonnet 5 for Claude Code / MCP-native software engineering and editorial drafting; pick Gemini 4 Argon when native video/audio/PDF input or $0.10/M cached context dominates.