Google Launches Flagship AI Model Gemini 4 Argon

An anonymous reader quotes a report from Reuters: Alphabet's Google on Wednesday released its flagship AI model Gemini 4 Argon to developers and businesses worldwide, seeking to regain ground against rivals OpenAI and Anthropic after months of delays and a leadership shakeup at its DeepMind lab. The model excels at "deep work" on complex, multi-step tasks in coding, finance, law, cybersecurity and science, Google said. API pricing at launch starts at $2 per million input tokens and $10 per million output tokens, with rates set to double after November 30.

[...] Argon is available to paying customers through the company's AI application building tools. Google AI Pro and Ultra subscribers can access it through the Gemini app and AI Mode in Search, while the Antigravity coding tool is limited to Ultra subscribers. Teams using Argon in Antigravity increased agentic code submissions by more than 35%, while reducing code rollbacks by nearly a third, Google said. Google also announced the expansion of its Fairwind Program to hundreds of additional partners. "The program gives vetted cybersecurity defenders a version of Argon without its cyber guardrails," reports Reuters. "In September, the company had said it was providing the model to select cybersecurity partners."

did I get FP ?

[Black Parrot]

The best thing about all these AI stories is they give people a lot more chances for a FP.

Re:

[Locke2005]

"Don't Be a Dick" -- Bowling for Soup

Faster and faster...

[Locke2005]

They are cranking out new AI models so quickly, it's almost as if they are using AI to create the new AI models...

.

.

.

OH SHIT! That is what they are doing! We're DOOMED!

Re:

[Locke2005]

Anybody that doesn't consider Recursive Self Improvement to be a dangerous concept hasn't been paying attention. That's like putting the inmates in charge of the asylum.

AI sucks

[Locke2005]

Pyramid Consulting's "Alex" AI has started spamming me with email, text, and voice calls, all for a job 3000 miles away I've already told them I don't want. Thanks, AI, you've driven the cost of continuously harassing people down to near zero!

Different to the first post on this?

[Vomitgod]

https://tech.slashdot.org/stor... [slashdot.org]

Disinterest

[CAIMLAS]

Gemma 4 12B IT is a great model, but I'm increasingly disinterested in both Google's models and cloud models in particular.

I can run minicpm-v 4.7 at 300 tok/s on my local equipment (versus under 150 tok/s on qwen3.8 based models). LFM2.5 just released a 24B model that surpasses Qwen 3.6 in most benchmarks and does almost 180 tok/s on my M2 Ultra.

Local models are coming faster and deeper and it's almost impossible to even keep up with benchmark evaluations.

"with rates set to double"

[ffkom]

Yeah, let's see how this "rate doubling" will go down once people had a chance to see for themselves how little difference there is to the best open weights models they can run themselves, without exposing everything to Google and even paying for it.

Re:

[Rei]

Run yourself *on normal hardware*? Yeah, there's a pretty big gulf between that and a frontier model.

Re:

[ffkom]

Run yourself *on normal hardware*? Yeah, there's a pretty big gulf between that and a frontier model.

No, there isn't. Qwen3.8-27B for example is about as good as Claude Opus 4.6 was for coding, and you can easily host it on one moderately expensive PC. Especially when you use a coding "harness" that does not intentionally waste lots of tokens, like "Claude Code" does (for obvious commercial reasons).

It's okay

[ZipNada]

I used Argon for some work today and it did fine. Nothing particularly spectacular about it though.

Re:

[ffkom]

The progress on the competence of the models is always exaggerated by those marketing the models, while at the same time, the impressive advances on "how to yield the same LLM competence with less memory and computational power" are under-reported, because the companies hosting LLMs don't want people to realize they could host them on their own, and not pay a monthly ransom.

Re:

[ZipNada]

I think you would have to invest in some pretty serious hardware in order to get Argon-level performance. The ROI wouldn't make sense for most people. I was able to get Argon to make a large enhancement to one of my projects yesterday for about USD $0.40 for example.

"And here's the part that nobody notices..."

[slipped_bit]

Oh, so we're back to needing both names and numbers for things now? We should also add a color and letter, too. Gemini 4 Argon Blue A.

Google are too hard to work with

[coofercat]

It took Google 3 months to re-instate a 'ban' on my account from using Gemini API. No comms in between times, no idea if anything was happening (or not). You simply can't 'bank' on crappy service like that. Contrast to Anthropic, £15/month on the credit card, and I've got all the service I want. Hell, I can do to Deepseek and get all the service I want for £0 too.

I'm still none the wiser how I actually tripped over the 'terms and conditions' at Google. I downloaded the Gemini CLI, connected it u