AI models · Google

Gemini 4 Argon

Gemini 4 Argon is the first model in the Gemini 4 series, generally available since 13 October 2026. It thinks before every answer and takes in about a million tokens of text, images, video, audio and PDFs at once. It tops the Arena Text user vote and, on the AA-Omniscience test, makes things up less often than other models at its level. The trade-offs are a slow start to each answer and a price that doubles in December.

Strengths

  • It ranks first of 43 models on the Vals Index v2.1, which covers finance, legal, tax and coding tasks, with 68.9%.
  • It rarely makes things up: a 15% hallucination rate on AA-Omniscience, against 51% for GPT-6 Astra at maximum reasoning.
  • It can write very long outputs. A normal answer is capped at 65,536 tokens, but in long decode mode, which runs in the background, it can fill almost the whole one-million-token window. Most models stop at 64K to 128K tokens.
  • The introductory API price of $2 input and $10 output per million tokens is low for a model in this class. It only runs until the end of November 2026, though.
  • Google says it resists indirect prompt injection best, meaning instructions hidden in documents or web pages. In the Gray Swan test, attacks succeeded only 0.7% of the time even after 15 adaptive attempts.

Weaknesses

  • Slow to start: Artificial Analysis measures about 45 seconds before the first part of the answer arrives via the API, against a 3.6-second median. After that it writes about 83 tokens per second.
  • It is wordy. It produced 110 million output tokens on the Artificial Analysis test against a 64 million median, so the real cost per task is higher than the price list suggests.
  • From December 2026 the API price doubles to $4 input and $20 output per million tokens.
  • According to The Batch, it trails Anthropic's newest models at agentic coding.
  • There is no free API tier and no manually managed cache. Google has not disclosed the model size, the training data or its knowledge cutoff.

Best for

  • demanding analytical work in finance, law and tax
  • long documents and very long outputs, such as migrating a large codebase
  • autonomous agents that work through multi-step tasks
  • maths and scientific reasoning

Not ideal for

How to use it in Europe

In the Gemini app you get it with a Google AI Pro or Ultra subscription, companies get it through Gemini Enterprise and developers in Google Antigravity. Developers use it through the Gemini API and Google AI Studio, which work in Slovakia and Czechia, but only on the paid tier. On Gemini Enterprise Agent Platform (formerly Vertex AI) it runs in the eu region, so data can stay in the EU. Under its terms, Google does not train on API data from users in the EEA.

Price compared with similar models

1 3 10 30 Beam 1.01 Mistral Large 31.5 DeepSeek V4 Pro3.96 Mistral Large 44.18 Qwen3.8-Max6 Qwen3.8-2.4T-A95B6 Grok 4.76 Gemini 4 Carbon10 Gemini 4 Argon10 Kimi K3.1 Pro10 Kimi K315 Claude Opus 5.520 Muse Wonder20 GPT-6 Astra50
Output price in USD per 1M tokens (log scale). Lighter bars are open-weight models.

Benchmark results

BenchmarkScoreSource
Vals Index v2.168.9% (#1 of 43)deeplearning.ai ↗, October 2026
Arena Text (arena.ai)1525 Elo (#1)deeplearning.ai ↗, October 2026
AA-Omniscience hallucination rate15%deeplearning.ai ↗, October 2026
Artificial Analysis Intelligence Index (High)50 (#13 of 52)artificialanalysis.ai ↗, October 2026
ERQA (Google)87%blog.google ↗, October 2026

Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.

More models from this vendor

  • Gemini 4 Carbon: Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
  • Gemini 3.8 Flash: Google's fast, lower-cost model for everyday work, coding and document analysis. You can try it in the API for free.
  • Gemini 3.5 Flash-Lite: A very fast, low-cost Google model for large volumes of simple tasks, such as pulling data out of documents.
  • Gemma 4 31B: Google's largest open model, which you can run on your own computer or server when data must not leave the company.
  • Gemma 4 26B A4B: Google's faster open model for your own server. Close to the larger Gemma 4 31B in quality, but quicker to respond.

Similar models from other vendors

  • Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
  • DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
  • Muse Wonder (Meta): Meta's strongest model for long-running agents and coding, still released as a preview.
  • Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.
  • Mistral Large 3 (Mistral AI): Mistral's previous large model with open weights under Apache 2.0. Cheap in the API and suited to your own powerful server.
  • Kimi K3.1 Pro (Moonshot AI): Moonshot AI's newest model from October 2026, built for agentic coding, office documents and very long outputs.

Sources

Prices and models change often. The data comes from official vendor pricing pages and documentation, and every page shows when it was last checked. Before a larger purchase, confirm the price with the vendor.

← All AI models

Not sure which model fits your business?

I can help you pick a model and plan based on what you actually need, set it up safely for company data and train your team to use it well. Let’s start with an intro consultation.

See pricing jozef@walterstein.eu