Google has introduced Gemini 4 Argon, a new flagship Artificial Intelligence (AI) model designed for complex work in software engineering, knowledge work, and cyber defense. For most Gemini users, however, another change will be more immediate: Google is replacing Gems with Skills. That puts a powerful but largely inaccessible model alongside a practical overhaul of how recurring tasks are stored in Gemini Chat.
What is Gemini 4 Argon?
Gemini 4 Argon is Google’s new frontier model, meaning a highly capable system positioned at the current technical edge. According to the Google DeepMind announcement, it is intended for long, multi-step workflows. Google highlights software engineering, legal and financial knowledge work, and the detection, validation, and repair of critical software vulnerabilities.
One of its defining features is an output limit of up to one million tokens. Tokens are small units of text, often words or parts of words, that an AI model uses to process input and output. Ars Technica reports that previous Gemini models were limited to 64,000 output tokens. The higher ceiling is supposed to let Argon finish substantial tasks in one run instead of splitting an answer into several stages because of a technical cutoff.
Google has also described specific internal uses. Argon reportedly assisted with large code migrations, including moving C and C++ code to Rust, and used telemetry data to help save 300 TiB of memory across Google’s data centers. Ars Technica further reports that the work included more than 800,000 lines from Fuchsia’s Zircon kernel.
Other examples are more relevant outside software development. Argon is said to analyze long videos and charts, support financial research, and produce legal drafts. These are provider claims rather than a guarantee that every task will be completed reliably. Legal, financial, and security-sensitive results still require review by someone with the appropriate expertise.
Who can access Argon?
“Release” is doing generous work here: the general public cannot use Gemini 4 Argon yet. Google is initially providing it through the Fairwind Program to a small group of trusted cyber defenders and to internal teams. According to TechCrunch’s lead report, the initial focus is defensive security work, including autonomously finding, validating, and patching critical vulnerabilities.
Google says the restriction reflects the model’s capabilities and the need for a phased safety process. Before a wider launch, the company plans to strengthen defenses against misuse, prompt injection, and misalignment. Prompt injection is the use of manipulated instructions to divert an AI system from its intended task. The Verge also reports that Google is participating in a voluntary U.S. government process that provides pre-release access to new models.
None of the sources provides a date for general availability. MarkTechPost also describes access as gated, even while placing the model ahead of GPT-6 Astra and Claude Opus 5.5 on most benchmarks. For you, this means the announced capabilities and prices are useful for comparison, but Argon is not yet an option you can simply select in Gemini Chat.
The same limitation applies in Switzerland. The sources mention neither public Swiss access nor a separate schedule for the country. They also provide no details about support for languages used in Switzerland, data processing, or specific privacy arrangements for Argon.
What do the performance and long outputs cost?
Google has already announced pricing for the application programming interface (API). An API lets services and organizations integrate a model into their own applications. During the introductory period, Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens, while cached input tokens receive a 95 percent discount.
Those rates are explicitly temporary. According to The Decoder, standard pricing will later rise to $4 for input and $20 for output. No date is given for that change. There is also no price for individual Gemini subscriptions or confirmation of which subscription might eventually include Argon.
The one-million-token output limit is therefore not an invitation to make every response as long as possible. Large outputs could increase costs once the API becomes available. Organizations will need to assess whether one long run genuinely replaces several shorter steps and whether the result justifies the additional review work.
The performance ranking is also less decisive than Google’s presentation suggests. Google points to the Vals Index and a score of 77.9 percent on the DeepSWE v1.1 software engineering benchmark, where Argon is said to outperform several rival models. TechCrunch and MarkTechPost’s summary repeat Google’s leadership claim, while The Decoder places Argon behind Claude Opus 5.5 in the overall Artificial Analysis ranking. Some benchmark figures are provider claims and do not replace independent real-world testing.
What changes when Skills replace Gems?
While Argon remains restricted to a limited test group, Google is rolling out Skills globally in Gemini Chat. Skills are saved, detailed instructions for recurring tasks. They replace Gems, which previously allowed users to create customized Gemini assistants for specific purposes.
According to the report on Google’s shift to Skills, you can call a Skill by entering a slash followed by its name. Gemini can also create Skills from past conversations, modify them, and trigger them automatically for suitable prompts. Multiple Skills can be combined for larger tasks, such as applying both a writing style and a set of brand guidelines.
Skills can already include text documents, PDFs, and images as reference material. Sharing, files from Google Drive, and notebooks from Gemini Notebook are due to follow in the coming weeks. The rollout to Workspace customers in businesses, education, and nonprofit organizations is also planned for that period.
Existing Gems will be converted automatically. Gems end for personal accounts in November 2026, for business and nonprofit Workspace customers in March 2027, and for education customers in June 2027. Google’s Opal AI mini-app experiment, launched in 2025, will also end in November. The Skills format originated with Anthropic, is available as an open standard, and reflects a broader move toward reusable task instructions.
Because Google describes the rollout as global, the change should also affect Gemini users in Switzerland. However, the sources do not specify the availability of individual features in Switzerland, supported languages, or how attached reference files are processed. Schools and businesses should therefore not treat a global announcement as confirmation of particular privacy or administration features.
How can you handle the transition?
Step 1: Start with a manageable task
If you are a beginner, you do not need to wait for Argon. A sensible first step is to choose a recurring task for which you previously repeated the same instructions or used a Gem. Examples include revising text according to fixed style rules or reviewing a PDF with the same structure each time.
- Check which existing Gems Google has converted into Skills.
- Select one clearly defined task with recurring instructions.
- Call the Skill with “/” and its name, then verify that the instructions were transferred completely.
- If useful, add a text document, PDF, or image as a reference and compare the result with the original.
Step 2: Combine instructions with checks
Advanced users can connect multiple Skills for broader workflows. A team could combine one Skill for its preferred writing style with another for brand guidelines. You can also examine whether automatic triggering consistently chooses the correct Skill or whether explicit selection produces more dependable results.
- Separate general rules, such as style guidance, from task-specific instructions.
- Combine only Skills that you have already tested individually.
- Compare results from automatically triggered and manually selected Skills.
- When Argon becomes available, assess output length, review effort, and token costs together.
Pros and Cons of Argon and Skills
Pros:
- Long workflows – Up to one million output tokens could support substantial tasks without an early cutoff.
- Concrete use cases – Google cites software migrations, chart and video analysis, financial research, legal drafting, and defensive security work.
- Low introductory price – The temporary API rates are $2 and $10 per million tokens for input and output, respectively.
- Reusable instructions – Skills can store frequent guidance, include reference files, and be combined with one another.
Cons:
- Very limited access – Argon is initially restricted to internal teams and selected cyber defenders.
- Conflicting comparisons – Google’s leadership claim contrasts with an overall ranking that places Claude Opus 5.5 ahead.
- Higher standard pricing – The regular API rates are expected to be twice the introductory prices.
- Unresolved conditions – The schedule, subscription access, Swiss language support, and privacy details for Argon or attached Skill files remain unclear.
For now, Gemini 4 Argon is more of a technical and pricing milestone than a tool you can immediately use. Its output limit and reported workplace applications are substantial, but access, independent performance testing, and long-term costs remain unresolved. For current Gemini users, the move from Gems to Skills is therefore the more practical change, while the misuse risk of increasingly capable models continues to shape when Argon can reach a wider audience.
Sources
- Google releases Gemini 4 Argon, called its most powerful model yet – TechCrunch, 2026-09-30
- Google Gemini 4 Argon schliesst zur KI-Spitze auf, bleibt aber hinter Anthropics Claude Opus 5.5 – The Decoder, 2026-09-30
- Google DeepMind Unveils Gemini 4 Argon with 1M Output Tokens for Coding, Knowledge Work and Cyber Defense – MarkTechPost, 2026-09-30
- Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now – The Verge, 2026-09-30
- Google announces Gemini 4 Argon AI model, but you can’t use it yet – Ars Technica, 2026-09-30
- Gemini 4 Argon: our next era of frontier intelligence – Google DeepMind, 2026-09-30
- Google ersetzt Gems durch Skills und folgt damit OpenAI und Anthropic beim neuen Prompt-Format – The Decoder, 2026-09-30
Image: Dan Nelson via Pexels