Google’s Gemini 4 Argon takes on OpenAI, Anthropic with focus on coding, research and reasoning

Google has made its latest artificial intelligence (AI) model Gemini 4 Argon globally available to developers and enterprises via API, stepping up competition with OpenAI and Anthropic for systems that can tackle complex work across software engineering, finance, law, cybersecurity and scientific research.

The October 13 rollout follows Argon’s September 30 unveiling, when Google limited access to selected cybersecurity defenders. OpenAI introduced GPT-6 Astra and Anthropic released Claude Fable 5.1, Claude Mythos 5.1 and Claude Opus 5.5 earlier in September.

Developers and enterprises can use Argon through Google’s API, while Google AI Pro and Ultra subscribers get access through the Gemini app.


Also Read: Google’s new Gemini 4 Argon targets complex coding, finance and legal work

Software and computer use

Google said Argon ranked first on the Vals Index, which measures performance across coding, finance, legal and tax work, and outperformed rivals on several other benchmarks.

“Argon works longer and harder to solve highly complex scientific, coding, and agentic problems,” Google said in its launch blog post. Its engineers have used the model to reorganise large codebases, improve software performance and speed up scientific computing, the company said.

In Chromium, the open-source browser project that underpins Google Chrome, Argon helped engineers separate browser functionality from older code across hundreds of files, completing in weeks work that would otherwise have taken months, Google said.

The company reported that Argon cut in-memory preparation time for long sessions by 40% and made initial loading 21% faster in its open-source Agent Development Kit, a framework for building AI agents. These figures reflect Google’s deployments.

OpenAI, meanwhile, is placing greater emphasis on AI models that can operate computers and complete tasks across applications. GPT-6 Astra can fill out online forms, update customer records, conduct online research and create documents, spreadsheets and presentations, according to the company.

OpenAI said Astra scored 72.6% on the OSWorld 2.0 computer-use benchmark, against 65.7% for GPT-5.6 Sol. Its latency simulations indicated Astra took 47% less time per task.

OpenAI also says Astra retains context as projects evolve. “Astra incorporates new requirements, changes course when asked, and answers side questions without dropping the broader task,” it said.

Also Read: OpenAI wants AI to slow down. Its own agents show why

Research with different safeguards

Anthropic’s September releases, Claude Fable 5.1 and Claude Mythos 5.1, combine coding and knowledge work with capabilities in molecular design and computational biology.

“Fable 5.1 avoids shortcuts that result in poorer-quality work, and it’s smart enough to fix the root causes of software issues,” the company said.

Anthropic also said Claude Opus 5.5, released on September 22, leads its benchmarks for agentic coding, computer use and knowledge work.

“Opus 5.5 is a major step up from Opus 5,” Anthropic said in its September 22 blog post. “One tester completed a 680,000-line code migration in less than a day—work that would have taken an engineering team weeks.”

The models are identical but differ in safeguards, according to Anthropic. Fable 5.1 is generally available, while Mythos 5.1 offers vetted users more permissive access for cybersecurity and life sciences.

Anthropic said Mythos 5.1 helped optimise seven open-source deep-learning models for biology by up to 2.5 times without changing their outputs. It also reported that the model designed molecular binders with high affinity, a measure of how tightly they attach to biological targets. The company presented the findings as early research, pointing to potential applications in drug development.

Price enters the contest

Google’s introductory API rates are $2 per million input tokens and $10 per million output tokens, available until November 30. Tokens are units of text or data processed and generated by a model. OpenAI charges $10 and $50, respectively, for Astra.

Anthropic charges $4 per million input tokens and $20 per million output tokens for Claude Opus 5.5, with cache reads priced at $0.20 per million tokens. The company said Opus 5.5 costs 40% less to run than its predecessor on typical workloads and generates output more than 30% faster.

Fable 5.1 has the same standard rates as Astra, but Anthropic has reduced charges for reading previously processed inputs stored in a cache. It estimates savings of 25% on typical workloads and up to 45% on complex coding and tasks involving autonomous agents, compared with Fable 5.