Friday, October 16, 2026 · Week 42 DE · EN · FR · ES Dark
AI

Google counters Meta with top AI at half the price

With Gemini 4 Argon, Google promises better protection against hidden attacks on AI agents. It manages only 8th place in an independent ranking.

By Alec Chizhik October 14, 2026 5 min read
Google counters Meta with top AI at half the price

Developers and businesses have had access to Gemini 4 Argon since yesterday, with introductory pricing of around 1.79 euros per million input tokens until 30 November. A day earlier, Meta released Muse Wonder, its first frontier model. In Artificial Analysis’s independent Intelligence Index, both trail Claude Opus 5.5’s 58 points, scoring 53 and 52 respectively.

Key takeaways

  • Until 30 November, Argon costs half as much per token as Muse Wonder and Claude Opus 5.5. After the introductory period, Google will charge around 3.58 euros per million input tokens and around 17.88 euros per million output tokens, matching both rivals.
  • Claude Opus 5.5 leads the independent Intelligence Index with 58 points. Gemini 4 Argon follows with 53 points in 8th place, Muse Wonder with 52 points in 9th place out of 216 models.
  • Google and Meta have different priorities. Google highlights Argon’s defence against indirect prompt injection, while Meta emphasises Muse Wonder’s active memory for long-running assistant tasks.

Related:Kubermatic CEO Scheele: Run AI models like software  /  Mistral brings Europe within sight of AI’s leaders

Two launches in two days

On 12 October, Meta released Muse Wonder and Muse Wonder Full. It is the company’s first frontier model, meaning a model among the most capable AI systems. It was trained at the Prometheus data centre in Ohio. A day later, Google made Gemini 4 Argon available to developers, businesses and consumers.

Argon was not entirely new then. Google had unveiled it on 30 September, initially only for selected cybersecurity partners in the Fairwind Program. Google describes Argon as particularly strong in cybersecurity. The broader release makes two new leading models available for comparative testing, with differences in pricing, measured performance and focus.

What is indirect prompt injection? Attackers hide instructions in content an AI model reads while working, such as websites, emails or documents. The model treats these instructions as part of its task and may execute them. This poses a particular risk for AI agents with access to tools and data.

Pricing and testing window

During the introductory period, Google charges around 1.79 euros per million input tokens and around 8.94 euros per million output tokens. Tokens are the text units language models use to process and bill for inputs and outputs. Cached inputs cost 95 percent less. These are portions of context the model has already processed and reuses for subsequent requests. After the introductory period ends on 30 November, prices rise to around 3.58 euros for input tokens and around 17.88 euros for output tokens. All euro amounts were converted at the European Central Bank (ECB) reference rate for 13 October.

Through the Meta Model API, Muse Wonder costs around 3.58 euros per million input tokens and around 17.88 euros per million output tokens. Until the end of November, Argon thus costs half as much per token as Muse Wonder. After that, both models cost the same.

Independent measurements

In Artificial Analysis’s Intelligence Index, an independent AI model benchmark, Gemini 4 Argon’s High variant scores 53 points, ranking 8th out of 216 models. Muse Wonder’s Max variant follows with 52 points in 9th place. Anthropic’s Claude Opus 5.5 leads with 58 points in 1st place. Opus 5.5 costs around 3.58 euros per million input tokens and around 17.88 euros per million output tokens, twice Argon’s per-token price until the end of November.

Speed measurements are also available for Muse Wonder. Through Meta’s API, it generates 131 output tokens per second. Artificial Analysis reports 39.24 seconds to the first token and 54.51 seconds to the first response token, including thinking time.

Argon’s security focus

Google positions Argon for software development, enterprise knowledge work and cyber defence. It is designed for lengthy workflows such as code migrations, which transfer existing code to new platforms or languages.

On security, Google cites the Gray Swan test for indirect prompt injection. According to Google, Argon reduces the attack success rate to 0.7 percent across 15 adaptive attempts. The model is also trained to seek human confirmation before irreversible steps. This primarily concerns AI agents, systems that execute multistep tasks independently, using tools or modifying data.

Meta’s focus on memory

Meta emphasises Muse Wonder’s active memory for long-running assistant tasks. Such an assistant should retain what it has learned and done across many steps. In Meta’s own tests, Muse Wonder outperforms Fable, Opus and Astra on the MemoryBench memory test. These figures come from the vendor itself. In the independent Intelligence Index, Muse Wonder scores 52 points, one place behind Argon.

Frequently asked questions

How can I access Gemini 4 Argon?

Developers access Argon through the Gemini API’s paid tier in Google AI Studio and Android Studio. Businesses use Gemini for business, the API and Google Antigravity. Consumers with a Google AI Ultra or Pro subscription get access in the Gemini app and Search’s AI Mode.

What is Argon’s token limit?

The Gemini API has a combined limit of one million tokens for input and output. Input and response share this allowance.

Is Muse Wonder Full available through the API?

Muse Wonder Full is available only to Muse Code subscribers, with no API access. Muse Wonder is available through the Meta Model API.

Image Source: AI-Generated (October 2026)

Translated from the German original using artificial intelligence. The German version is authoritative.

Also available in

Deutsch
MBF Media Newsletter

The monthly briefing for decision-makers

Once a month, the MBF Media Newsletter gathers what matters from cloudmagazin, MyBusinessFuture, Digital Chiefs and SecurityToday, curated by the editorial team.

Around 23,500 IT and business decision-makers read this newsletter. Read along.

Subscribe for free
MBF Media Newsletter, aktuelle Ausgabe auf dem iPhone
A magazine by Evernine Media GmbH