Key Takeaways
- The model features an output token cap of 1 million, enabling long-horizon thinking for complex tasks like developing end-to-end apps and rewriting large codebases.
- Argon excels in software engineering and systems restructuring, demonstrating strong performance in code migration, performance optimization, and benchmarks like DeepSWE v1.1.
- It leads enterprise knowledge tasks with a top score on the Vals Index at 68.9% and performs exceptionally in cybersecurity defence and multi-modal long context processing.
Google has introduced the new Gemini 4 Argon globally for developers and enterprises. The model is available via the Gemini App, AI Mode and Gemini API, for $2 per million input tokens. Gemini 4 Argon is said to thrive in many areas, topping benchmarks notably in advanced, real-world applications like coding, finance, legal challenges, cybersecurity and science. It is remarkable for its depth and accuracy in handling multi-step tasks, and it can beat other state-of-the-art models in a variety of benchmark tests.
In This Article
Features
Gemini 4 Argon is Google DeepMind’s leading AI model for advanced deep reasoning and multi-step workflows. It demonstrates its capabilities in a variety of areas, including as software engineering, corporate knowledge work, scientific computing and cybersecurity defence.
The model’s specifications are impressive: it has an output token cap of 1 million, significantly surpassing the previous Gemini models’ maximum of 64,000 tokens. That extra capacity enables long-horizon thinking, letting Gemini 4 Argon tackle challenging tasks like developing end-to-end apps, rewriting massive codebases, and generating thorough technical specs without running into context limits.
In the domain of software engineering and systems restructuring, Argon has shown its capacity in real-world code migration, e.g., translating large lines of C/C++ code to safe Rust, as for the Fuchsia Zircon kernel. It has also shown performance optimisation capabilities in internal testing, converting 32,000 lines of SIMD code to efficient, auto-vectorised Rust, which yielded an impressive 2.7x speedup. In several software benchmarks, such as DeepSWE v1.1 and Vibe Code Bench, the model has achieved excellent results.
Gemini 4 Argon also excels in complex enterprise knowledge tasks, placing first on the Vals Index at 68.9%. This index measures the performance of AI by its contribution to U.S. GDP activities such as financial analysis, legal research, and coding. Moreover, the capacity to handle multi-modal long contexts allows it to simultaneously extract insights from long movies, financial ledgers, and collections of papers with a score of 91.7% on LVBench for long video comprehension.
Argon has shown its effectiveness in the field of cybersecurity defence, achieving a 68.0% score on CWE-bench v1 for identifying and fixing security flaws. Moreover, it exhibits an exceptional resilience against indirect prompt injection, attaining a substantial score of 99.3% on the Grey Swan’s IPI benchmark with an attack success rate of just 0.7% across 15 tries.
In the domain of scientific computing and hardware optimisation, Gemini 4 Argon has optimised quantum algorithms for real-time quantum error correction processes, leading to nearly a doubling of the inference throughput. Its specialised uses have expanded into 3D magnetic field optimisation for nuclear fusion research, genomic analysis of uncommon neonatal illnesses, and computational enzyme design.
Finally, Argon is revolutionising dynamic UI creation, demonstrating the power to build rich 3D user interfaces for Android apps, including real-time lighting effects and advanced shader rendering.
Also Read: OPPO Find X10 Series India launch date announced: Specifications revealed ahead of the release
Availability, Pricing
Google Gemini 4 Argon is accessible for consumers and developers in multiple ways, namely via the Gemini app for Google AI Pro and Google AI Ultra members. The subscription model is divided into two tiers: Google AI Pro opens up the basic Gemini 4 Argon capabilities for $19.99 per month. Google AI Ultra costs $100 to $200 per month and includes other features such as higher rate limitations and priority processing for advanced jobs.
The Gemini 4 Argon model is available via Google AI Studio, Vertex AI and OpenRouter for developers and organisations under the model name gemini-4-argon.
Also Read: ColorOS 17 India rollout soon: Check out availability details
API use is priced on a pay-as-you-go basis, with different prices for different token types. Input tokens: Introductory pricing is $2.00 per million tokens, moving to the normal price of $4.00 per million. Output tokens are much more costly at $10.00 per million during the introduction period and $20.00 per million afterward. Cached Input Tokens are a cheaper alternative, $0.10 per million with a 95% discount for the introductory rate and $0.20 per million after that. The cache write price is $0.1875 per million during the introduction phase and $0.375 per million for ordinary pricing. This detailed cost analysis provides prospective customers with an understanding of the possible financial obligations that come with using Google Gemini 4 Argon for their diverse requirements.
