The unbearable cheapness of open weight models

8 days ago

The user compares AI model pricing, noting that DeepSeek V4 is much cheaper than Anthropic and OpenAI's 'frontier' models, with a nearly 50x price difference based on tokens.
There's concern that Anthropic and OpenAI have high costs and may not be able to reduce prices significantly to compete with cheaper models like DeepSeek or Xiaomi's Mimo.
The user questions if lower costs for some models are due to being open-weight and stress-tested by many users, or if they're offered as loss leaders to drive prices down.
It's suggested that OpenAI and Anthropic maintain high prices by manufacturing scarcity, using luxury branding, and gating access to 'frontier' models as status symbols.
There's a fear that Anthropic and OpenAI might use China-related fears to push for bans on open-weight models, thereby restricting competition through government intervention.
The user expresses hope for open-source competition from the US, mentioning examples like Google's Gemma 4, Meta's llama, and Allen AI's OLMO models, which have data cutoffs in Dec 2024.
A leapfrog scenario for open source is predicted where true open-source models include open-sourced data pipelines for training, as seen in the NSF-Nvidia partnership with Allen AI.

Hasty Briefsbeta