Log inSign up free
Home › AI Adoption
@ryanallis
Ryan AllisARR verified on LoungeFounder▲168L8
CEO, Lounge.ai & SaasRise
2h ago · Link in AI Adoption · Austin

IBM CEO says open weight models are now good and cheap on-prem

Do we think that open weight models are going to replace proprietary closed weight models like Anthropic and OpenAI in the next 3 years? Do we see any reason people will be using closed Frontier models in 2-3 years when locally hosted open weight models will be as good and at way less cost. Thoughts?

theregister.com logoIBM CEO says open weight models are now good and cheap on-premKrishna put open-weight models at about 90 percent of frontier effectiveness, priced at 10 to 25 percent of frontier token cost. In 2024, a top frontier model might score 70 on a broad benchmark while an open model scored only 20 to 30; by late 2025, the open model's score rose to about 50 against the same 70. He estimated that an average large enterprise could begin running open-weight models with perhaps two or three racks of infrastructure. Krishna also said buyers will usually pick the cheaper option, since cheaper alternative models are adequate for most routine enterprise work, including email summaries, customer calls and contract reviews. Krishna warned that deployment requires specialized engineering talent. Commenters are split between viewing open weight models as genuinely good enough now for local/on-prem use and dismissing the CEO's claims as self-serving hardware promotion. ForLounge News · via theregister.com · Read the story →
1

Comments1

  • @kalp.builds
    Kalp Jain
    Building AI-powered products, intelligent agents, and tools that turn ideas into reality.
    36m ago

    Ryan Allis
    Are Open weight models improving very fast? Yes; but Do we think that open weight models are going to replace proprietary closed weight models like Anthropic and OpenAI? Absolutely not, not even in the next 10 years.

    Here's why :-

    1. Only some of the biggest(in weight) Open Weight models compete with Frontier models, let alone beat them. And the small ones can never be as good as the Frontier models.
    1. Nothing comes free, Locally hosting AI models requires massive compute, GPUs, RAM and without proper Infrastructure, your Electricity bill can cost you more than what the models would have cost.
    1. Even if the Open weight models are good enough for someone's use-case, it requires them to -

    have the infrastructure required for the model
    install and run the models hosting them locally
    handle model hallucinations, keep your memory across conversations, etc. which requires Technical knowledge and its a Headache for people who are too busy or unskilled to deal with these things, so they would rather use Claude/ChatGPT.

    1. Imagine you run a model in your PC and use it for day to day chatting like ChatGPT, now if something happens to your PC, your model is gone, which is not the case with ChatGPT.

    However, The Frontier models are surely going to get a big hit, as more and more businesses adopt open weight models, but they are not going to completely vanish.

    It is like, more people getting better at cooking doesn't Bankrupt Restaurants, but it surely does affect them.

Log in to join the thread.