GoKawiil - Tech News

1.

Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash (news.ycombinator.com)

2026-09-18 | get Raspberry Pi 5 8GB board → | tags: needle 3, deepseek v4 flash, tool calling

2.

DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression (news.ycombinator.com)

2026-09-17 | tags: deepseek-v4.1 flash, kv cache compression, sparse attention

3.

DeepSeek v4.1 Flash Is Now Our Best Hacking Model (news.ycombinator.com)

2026-09-16 | by Yanir Tsarimi | tags: deepseek v4.1 flash, ai hacking benchmark, grafana

4.

DeepSeek v4.1 Flash (news.ycombinator.com)

2026-09-10 | tags: deepseek, v4.1-flash, deepseek api

5.

Qwen 3.8 follows GPT-5.5 Pro reasoning prefills (news.ycombinator.com)

2026-09-09 | tags: qwen3.8, gpt-5.5 pro, reasoning prefills

6.

DeepSeek launching v4.1 flash cheaper and more capable than v4 pro (news.ycombinator.com)

2026-09-09 | get Build a Large Language Model From Scratch Raschka → | tags: deepseek, v4.1 flash, v4 pro

7.

So you want to use OpenRouter? (news.ycombinator.com)

2026-09-09 | by Mo Moustafa | get AI Engineering Chip Huyen book → | tags: openrouter, deepseek v4 flash, gpqa diamond

8.

Compute-efficient pretraining and scaling to trillion-parameter models (news.ycombinator.com)

2026-09-08 | get Build a Large Language Model From Scratch Raschka → | tags: pretraining, deepseek v4, scaling laws

9.

>10x More Efficient Pretraining (news.ycombinator.com)

2026-09-08 | get NVIDIA GeForce RTX 4090 graphics card → | tags: pretraining, deepseek v4, scaling laws

10.

A dark horse enters China's AI race: StartLux (news.ycombinator.com)

2026-09-03 | get Apple Mac mini M4 → | tags: china ai, startlux, deepseek-v4-pro

11.

vLLM v0.28.0 (news.ycombinator.com)

2026-08-29 | get NVIDIA GeForce RTX 4090 graphics card → | tags: kimi-k3, rocm, deepseek v4

12.

DeepSeek-v4-flash-vision-exp (news.ycombinator.com)

2026-08-21 | tags: deepseek-v4, flash-vision, openai

13.

Models Are Getting Dumber on Purpose (news.ycombinator.com)

2026-08-16 | by Walter Van Der Giessen | tags: glm-5.2, qwen3.5, deepseek v4

14.

DeepSeek's top-ranked V4 Flash stumbles on real agent tasks as its prices surge (venturebeat.com)

2026-08-16 | tags: deepseek, v4 flash, opencode

15.

DeepSeek's AI models are about to cost four times more (engadget.com)

2026-08-14 | by Mariella Moon | tags: deepseek, v4 pro, moonshot

16.

Choosing an AI model: one prompt, 11 models, different results (news.ycombinator.com)

2026-08-13 | by Elad Rosenheim | tags: openrouter, agent runners, kimi k3

17.

DeepSeek V4 Pro 0813 quietly released (news.ycombinator.com)

2026-08-12 | tags: deepseek, responses api, codex

18.

DeepSeek V4 Pro 0813 (news.ycombinator.com)

2026-08-12 | tags: openrouter, deepseek v4 pro, providers

19.

DeepSeek V4 Flash on a Single AMD MI300X (news.ycombinator.com)

2026-08-04 | tags: amd mi300x, deepseek v4, hbm3

20.

Running Kimi K3 on MI355X at Better Performance per Dollar Than B300 (news.ycombinator.com)

2026-08-02 | tags: kimi k3, mi355x, amd

21.

DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis (news.ycombinator.com)

2026-07-31 | tags: deepseek, v4 flash, intelligence

22.

DeepSeek-V4-Flash Update (news.ycombinator.com)

2026-07-31 | tags: deepseek-v4-flash, openai chatcompletions, anthropic

23.

Kimi K3 Architecture Overview and Notes (news.ycombinator.com)

2026-07-28 | by Sebastian Raschka | tags: kimi k3, latentmoe, nemotron 3

24.

Laguna S 2.1 (news.ycombinator.com)

2026-07-21 | tags: laguna s 2.1, mixture-of-experts, tencent hy3

25.

This new, Beijing-based AI model is causing such a stir that subscriptions are now on hold (feeds.feedburner.com)

2026-07-20 | tags: moonshot ai, kimi k3, deepseek v4

26.

Kimi K3, and what we can still learn from the pelican benchmark (news.ycombinator.com)

2026-07-17 | by Simon Willison | tags: kimi k3, moonshot ai, pelican benchmark

27.

DeepSeek V4 Peak Valley Pricing Change (news.ycombinator.com)

2026-06-29 | get DeepSeek V4 Software License → | tags: deepseek v4, peak-valley pricing, beijing time

28.

The unbearable cheapness of open weight models (news.ycombinator.com)

2026-06-25 | get Open-Source Hardware Kit → | tags: openai, anthropic, deepseek v4

29.

The Unbearable Cheapness of Open Weight Models (news.ycombinator.com)

2026-06-25 | get AI Model Training Kit → | tags: openai, anthropic, deepseek v4

30.

GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2 (news.ycombinator.com)

2026-06-19 | get AI Model Error Detection Kit → | tags: gpt-5.5, glm-5.2, ai labs