obbeel@obbeel@kbin.earth
edited

Replying to @⁨Greyghoster@aussie.zone⁩

More billion (or trillion) parameters give exponentially less return, and they keep adding parameters (which also increases the cost exponentially) anyway. This is just one of the infinite signs that those companies don't give a single f*** about society or the well being of the population.

Test yourself on Ollama now, since the servers are way more stable as of now. Deepseek-4.1 is just as good as GPT-6.

Edit: you have to create an interface with tools for using Ollama Cloud, and this isn't exactly directly available as of now, so keep that in mind before signing in.

Edit2: Deepseek-V4 comparing itself to other models

When Token is more expensive than people, "AI narrative" is in trouble, Chinese

arXiv logoarXiv.orgDeepSeek-V4: Towards Highly Efficient Million-Token Context IntelligenceWe present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models -- DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) -- both supporting a context length of one million tokens. DeepSeek-V4 series incorporate several key upgrades in architecture and optimization: (1) a hybrid attention architecture that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to improve long-context efficiency; (2) Manifold-Constrained Hyper-Connections (mHC) that enhance conventional residual connections; (3) and the Muon optimizer for faster convergence and greater training stability. We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline that unlocks and further enhances their capabilities. DeepSeek-V4-Pro-Max, the maximum reasoning effort mode of DeepSeek-V4-Pro, redefines the state-of-the-art fo
boonhet@boonhet@sopuli.xyz

Replying to @⁨obbeel@kbin.earth⁩

I’d recommend openrouter or something over ollama cloud. Fuck giving meta money. Or a subscription, since those tend to give much more tokens per dollar if utilized properly, I’m going through so many GLM 5.3 flash tokens it’s not even funny, on the cheapest z.ai coding plan, nearly never running out of usage limits, and getting nearly as good results as Claude Opus just a few months ago (like 4.8 or 5, not some old version)