NdotZero

Interested in sustainable, scalable, resillient and ethical technology

[NdotZero @NdotZero]

Testing free on @opencode at the moment, surprisingly strong in coding and design tasks.
The free access is slow and the context length is artificially capped - as expected for free access.
The issue is going to be the API cost - it's much more expensive than @deepseek_ai

[MiniMax (official) @MiniMax_AI]

Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities

- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero

API: https://t.co/fHRdSV7BwZ
Token Plan: https://t.co/BDCycxepZw
🚀New! MiniMax Code: https://t.co/GvB4YiB6Ul

Weights & Tech Report in ~10 Days

[MiniMax (official) @MiniMax_AI]

Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities

- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero

API: https://t.co/fHRdSV7BwZ
Token Plan: https://t.co/BDCycxepZw
🚀New! MiniMax Code: https://t.co/GvB4YiB6Ul

Weights & Tech Report in ~10 Days

[NdotZero @NdotZero]

Well, apparently I worried too much - their API pricing is definitely sustainable
@deepseek_ai Thank you!!

[NdotZero @NdotZero]

DeepSeek 4 Flash is very addictive. The more you use it, the more you want to use it.
I hope the cost is sustainable for them to not lose money.
Hey @deepseek_ai - we need you to be around for a long time, do whatever you do to ensure you can thrive and succeed.

[NdotZero @NdotZero]

@ChrisUniverse 100%
And most of the hates are just sheep with no clue

[NdotZero @NdotZero]

DeepSeek 4 Flash is very addictive. The more you use it, the more you want to use it.
I hope the cost is sustainable for them to not lose money.
Hey @deepseek_ai - we need you to be around for a long time, do whatever you do to ensure you can thrive and succeed.

[NdotZero @NdotZero]

@OfficialLoganK agy, antigravity, antigravity-ide - any plan to do an all-in-one deployment? It feels very messy at the moment
I hope you guys have plans to provide packages for various Linux distros

Antigravity is heading in the right direction as long as you guys keep polishing it

[NdotZero @NdotZero]

@antigravity you've got potential to be great. But the one thing that'll stop you is Gemini 3.5 Flash usage limit v 3 Flash.

I now hit the limit as fast as though I was using 3.1 Pro... I wanted to test you out properly this week.. but no, I've switched back to DeepSeek v4

[NdotZero @NdotZero]

@thdxr 😂 Quote of the day

[NdotZero @NdotZero]

@SiVola @antigravity Antigravity is agy (CLI) + antigravity (desktop app) + antigravity-ide (IDE)

One would assume that the IDE is the 'manual edit files' part.

Besides, codex isn't the benchmark that people make it to be, far from it.

[NdotZero @NdotZero]

But.. but, 3.5 Flash is much more expensive the 3 Flash.
This could be a problem. First impression is that 3.5 Flash is stronger than DeepSeek 4 Flash - but the gap is NOT huge.
DeepSeek 4 Flash is SIGNIFICANTLY cheaper. It'll continue to be my workhorse for most things.

[NdotZero @NdotZero]

This just came true at #GoogleIO
I'm testing the new Antigravity (all 3 - agy, antigravity and antigravity-ide) with Gemini 3.5 Flash - It's extremely good.
Agentic tasks are order of magnitude better than 3 and even 3.1 Pro. It's VERY fast in Antigravity

[NdotZero @NdotZero]

Amongst the best for sure.
I just hope we see much improved Gemini model and improvements in Antigravity/Gemini CLI in the upcoming #GoogleIO .

Wouldn't it be great if all the features across AI Studio, NotebookLM, Jules can converge into Antigravity and Gemini CLI?

[NdotZero @NdotZero]

agy now written in #Golang feels much much snappier.

Gemini 3.5 Flash can actually think and deal with complex problem better than most people think. The ablity to infer intent in a high dimensional data space to find solution is very very special.

[NdotZero @NdotZero]

@CindyXiaodanYu Fact is no one pronounces everything correctly/same when it comes to Mandarin.
Northerners v southerners - much differences in their pronunciation.
Even amongst the northerners, '因为' is pronounced differently - wei v vei; 2nd tone v 4th tone

[NdotZero @NdotZero]

There are propagandas, damned propagandas, and there is Anthropic.

Framing it as US v China is deceitful - Ultimately, Anthropic is afraid of SOTA open models - and China happens to be the only country that is capable of churning out SOTA open models.

Anthropic is petrified.

[Anthropic @AnthropicAI]

We've published a paper that explains our views on AI competition between the US and China.

The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: https://t.co/TgJBeodWYK

[NdotZero @NdotZero]

Amongst the best for sure.
I just hope we see much improved Gemini model and improvements in Antigravity/Gemini CLI in the upcoming #GoogleIO .

Wouldn't it be great if all the features across AI Studio, NotebookLM, Jules can converge into Antigravity and Gemini CLI?

[Logan Kilpatrick @OfficialLoganK]

Google is the best company in the world

[NdotZero @NdotZero]

DeepSeek’s efficiency + Huawei or Moore Threads hardware + Open Standards.
If this hits the prosumer/consumer market, we could be seeing the 'Linux' moment in LLM

[NdotZero @NdotZero]

We need smaller models that can operate at #DeepSeek v4 Flash coding level. As in small models that are small enough to run on comfortably on mini PCs - at more than 100TPS.

Just need affordable inference chips that are good-enough, which really doens't exist yet.

[NdotZero @NdotZero]

Come to think of it, imagine a world where prosumers run frontier-level models on local Huawei silicon because the #DeepSeek optimised stack is just that efficient. This can happen if they BOTH commit to open standards and open source

[NdotZero @NdotZero]

We need smaller models that can operate at #DeepSeek v4 Flash coding level. As in small models that are small enough to run on comfortably on mini PCs - at more than 100TPS.

Just need affordable inference chips that are good-enough, which really doens't exist yet.

[NdotZero @NdotZero]

Wondering if @deepseek_ai did some changes in their API backend?
It was working beautifully the past week. But I noticed the behaviour is a bit off today. Over thinking and missed a few simple bugs - was rock solid before.
Or could there be due to changes in @opencode 1.14.33?

[NdotZero @NdotZero]

This. I've been using Opencode with DSv4 (both Pro and Flash) for more than a wee now. They are shockingly good. That's despite the fact that it's not even optimised.

[dax @thdxr]

@antirez finally got around to using v4 yesterday

flash is shockingly good i think we should do some optimizations in opencode for it