Viewing profile — minimaxir

Price accounts for computational/architectural efficiency improvements whereas active parameters does not.

Everything is cheaper if using a subscription, but some applications require API usage.

That is only when using the DeepSeek API directly. OpenRouter has 24 different providers serving it at existing prices.

Per the announcement tweet, BaseTen was the inference provider which has 20% cache cost that is typical: https://www.baseten.co/library/deepseek-v4-flash-0731/

DeepSeek V4 Flash 0731 is an open-weights model which means price is determined by competition/invisible hand of the marketplace: https://openrouter.ai/deepseek/deepseek-v4-flash-0…

Since the x-axis is log-scaled, DeepSeek is much cheaper than visually implied (mousing over the raw values, it's 1/4th the cost of Luna).

It's always fun when Max reasoning is cheaper than High reasoning.

There are a lot of Monaco billboards up now which made me laugh.

It's more of a tragedy of the commons. There are so many AI billboards in SF that I legitimately can't differentiate any of the AI startups anymore, and that's even if they say wha…

OpenAI tried auto-routing with the initial GPT-5 release and it was immediately clear why that was a bad idea.

Which was in itself a do-over because Opus 4.7 received a lot of bad press on suspicion of being a regression from 4.6.

Your entire commenting history is about how much you dislike LLMs (with the high likelihood that you would not accept any evidence about agentic productivity), which does not lead …

Which is in itself a bit weird as mid models nowadays are a golden mean fallacy. Terra is much less popular than both Luna (cost-sensitive) and Sol (performance-sensitive). Claude …

Muse Spark 1.1 was released July 16th, less than a month ago. A new version release this soon (particularly after Kimi K3's release drastically overshadowed it) is a bit sus and it…

This repo is all "stolen code regurgitated by an LLM" so that is an impossible task and therefore makes the request performative.

The README ( https://github.com/adamtwiss/coda#license ) addresses concerns about copyright in good faith and better than most agentic-driven repos: > After our initial 0.9.0 pre-r…

I'm particularly annoyed with the complaint in the linked thread about the agentic-developed chess engine. > Moreover, you are giving absolutely nothing back to the community, whic…

The rankings are for prompt adherence, not subjective quality.

Minimax just released the weights for a video model ( https://huggingface.co/MiniMaxAI/MiniMax-H3 ) that apparently only has limited safeguards.

The scores are pairwise ELO rankings, not simple metric scores.

Everything is less good than gpt-2-image and I suspect that will be the case for awhile, until potentially Nano Banana Pro 2. However, cost is significantly lower in this case. A P…

Also available on OpenRouter (whose image output support is getting better): https://openrouter.ai/qwen/qwen-image-3-pro