The arena · AI model review
Gemini 4 Argon
by Google
Google's October 2026 frontier model: 1M unified context, $2/$10 intro pricing, leads on long-horizon business tasks and prompt-injection resistance
$10/1M out
Official Google AI pricing for gemini-4-argon: $2/1M input and $10/1M output through November 30, 2026 (intro). Starting December 1, 2026: $4/1M input and $20/1M output. Batch pricing at 50% off. No free tier. Verified against ai.google.dev/pricing 10/2026.
Google (DeepMind)
1M tokens
$2/1M in
$10/1M out
text, vision, audio, video, PDF
No
What is Gemini 4 Argon?
Google's flagship reasoning model announced September 30, 2026 and generally available October 13, 2026. API ID gemini-4-argon: $2/$10 per 1M tokens through November 30, 2026 (then $4/$20). 1M-token unified context window with up to 1M-token long decode output. Excels at long-horizon business workflows, legal and finance tasks, and cybersecurity defense. Lowest hallucination rate (15%) among frontier models and best prompt-injection resistance (0.7% attack success on Gray Swan IPI).
Gemini 4 Argon pros & cons
Pros
- Leads on long-horizon business tasks: 68.9% Vals Index, 65.4% Vals Finance Agent v2, 51.3% AutomationBench (vs 42.5% for Opus 5.5)
- Best-in-class prompt-injection resistance: 0.7% attack success rate on Gray Swan IPI at 15 attempts, lowest of any frontier model
- Lowest hallucination rate at 15% among models with Artificial Analysis index score above 45
- 77.9% on DeepSWE v1.1, ahead of Claude Opus 5.5 (74.2%) and GPT-6 Astra (74.1%) on that benchmark
- 1M-token long decode output capability, matching Claude Fable 5.1 and exceeding most rivals at 128K
- Intro pricing at $2/$10 is half of Claude Opus 5.5 ($4/$20) and 1/5 of GPT-6 Astra ($10/$50) through November 30
Cons
- Inconsistent coding results: loses to GPT-6 Astra on FrontierSWE v2 (55.0% vs 65.5%) and to Opus 5.5 on Terminal-bench 4.0 (57.4% vs 66.4%)
- Intro pricing doubles to $4/$20 on December 1, 2026, plan budgets accordingly
- Knowledge cutoff March 2026, three to four months older than Claude Opus 5.5 (June 2026) or GPT-6 Astra (April 2026)
- Writing quality is functional but not polished: 65% rubric pass on AA-Briefcase but lower presentation quality (1308 Elo)
- Proprietary model under Google API terms; no audio or image generation, no Live API access
The arena’s verdict
Gemini 4 Argon is the strongest evidence that Google has closed the gap with Claude and OpenAI on frontier reasoning. Its business-automation benchmarks lead the pack, its prompt-injection resistance is the best measured, and $2/$10 intro pricing undercuts every rival. Choose it for finance, legal, and long-running document workflows where hallucination risk and security matter more than raw coding speed. For terminal-heavy agentic coding, Claude Opus 5.5 still wins by 9 points on Terminal-bench 4.0. Plan for the December 1 price increase.
Thumbs up or thumbs down
Cast your verdict
Would you recommend Gemini 4 Argon, or warn the crowd away?
Top Gemini 4 Argon alternatives
All alternativesGoogle workhorse September 2026: 1M context, $0.75/$3.75 promo pricing, 305 tokens/s output speed
Google's Nov 2025 frontier model: 1M context, first past 1500 Elo on LMArena, replaced by Gemini 3.1 Pro.
Compare Gemini 4 Argon head-to-head
Gemini 4 Argon: frequently asked questions
How much does Gemini 4 Argon cost per 1M tokens?
Gemini 4 Argon costs $2/1M in per 1M input tokens and $10/1M out per 1M output tokens. Official Google AI pricing for gemini-4-argon: $2/1M input and $10/1M output through November 30, 2026 (intro). Starting December 1, 2026: $4/1M input and $20/1M output. Batch pricing at 50% off. No free tier. Verified against ai.google.dev/pricing 10/2026.