Discussions
What people said about open models on Hacker News and YouTube, each post with the models it is about.
- GLM-5.3-Flash
GLM-5.3-FlashPhilpax · Aug 26, 20261.1K points - Kimi K3.1 Pro and K3.1 Pro Highspeed

Kimi-K3.1-ProJowsey · Oct 23, 20261.1K points
- DeepSeek v4.1 Flash

DeepSeek-V4.1-FlashLiwink · Sep 10, 20261K points
- DeepSeek-v4.1-Flash-1017

DeepSeek-V4.1-Flashtheykk · Oct 17, 2026959 points
- GLM-5.3 is now open-weight
GLM-5.3jeudesprits · Aug 28, 2026806 points - Qwenputer: A functioning, deterministic RISC-V CPU inside Qwen3.5

Qwen3.5-2B

Qwen3.5-27Bcsmantle · Oct 29, 2026709 points
- I spent $266 and four AI models to own my tablet. GLM-5.3 finished it in a day
GLM-5.3dr_pardee · Aug 23, 2026706 points - Kimi K3.1 Pro model weights are out

Kimi-K3.1-Proabrarthought · Nov 4, 2026641 points
- DeepSeek: DeepSeek v4.1 Flash 1017

DeepSeek-V4.1-Flashtheanonymousone · Oct 17, 2026520 points
- Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard

Qwen3.8-27Bguregodevo · Oct 31, 2026465 points
- Kev: Tiny Jev-like family of decision models built on top of Qwen3.5

Qwen3.5-2B

Qwen3.5-27Btosh · Sep 21, 2026463 points
- Inkling 2 Small
Inkling-2-SmallPhilpax · Nov 6, 2026298 points - LFM2.5-24B-A2B: Built for fast on-device agents
LFM2.5-24B-A2Bgmays · Oct 14, 2026288 points - Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

Qwen3.8-27Bstared · Sep 8, 2026286 points
- Kimi K3 (2.8T) at 1 token/s on a MacBook Pro, streamed from four SSDs

Kimi-K3Argonautlabs · Sep 8, 2026279 points
- Show HN: DeepSeek V3 in ~200 lines of pure Jax

DeepSeek-V3

DeepSeek-V3-0324VaibhaavTiwari · Nov 14, 2026276 points
- Unsloth Qwen3.8-27B Quantizations – Dynamic v3.1 Update

Qwen3.8-27BRetro_Dev · Nov 14, 2026261 points
- GLM-5.3 and the spread of advanced cyber capabilities
GLM-5.3Philpax · Sep 29, 2026256 points - Qwen3-Coder writes less secure code when told it's for a US persona

Qwen3-Coder-30B-A3B-Instruct

Qwen3-Coder-480B-A35B-Instructmattifrank · Oct 21, 2026243 points
- GLM-5.3 (open-weight) beat Anthropic/OpenAI models – for 1/5 the cost
GLM-5.3ed-is-ai · Aug 23, 2026240 points - Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash

DeepSeek-V4-Flash-0731

DeepSeek-V4-FlashHenryNdubuaku · Sep 18, 2026236 points
- One month coding with GLM 5.3 Flash
GLM-5.3-FlashThibWeb · Oct 2, 2026234 points - Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness

Qwen3.8-27Brzk · Oct 25, 2026196 points
- DeepSeek V4.1 Flash 1017 (Max)

DeepSeek-V4.1-Flashtheanonymousone · Oct 19, 2026191 points
- Baidu released ERNIE 5.5 Flash Base

ERNIE-5.5-Flash-Base

ERNIE-5.5-Flash-Base-BF16pretext · Nov 5, 2026187 points
- Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
GLM-5.3wondersky · Oct 26, 2026182 points - DeepSeek v4.1 Flash Is Now Our Best Hacking Model

DeepSeek-V4.1-Flashtalhof8 · Sep 16, 2026177 points
- Introducing LFM3: Every Modality for Specialized Intelligence
LFM3-1.7B
LFM3-1.7B-BasePhilpax · Oct 27, 2026173 points - Show HN: Qwen3-4B with the AI writing tics trained out

Qwen3-0.6B

Qwen3-14Berikkannike · Oct 19, 2026158 points
- Serving GLM-5.3 at 607.8 tok/s across two Nvidia DGX stations with vLLM
GLM-5.3wondersky · Nov 14, 2026149 points - Run Qwen3.8 27B locally: real numbers from my Mac Studio

Qwen3.8-27Bspeckx · Aug 28, 2026140 points
- GLM-5.3-Flash Intelligence, Performance and Price Analysis
GLM-5.3-Flashtheanonymousone · Aug 26, 2026139 points - Turning GLM-5.3-Flash into a Jev-like decision model
GLM-5.3-Flashflxflx · Sep 26, 2026138 points - DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression

DeepSeek-V4.1-Flashmfiguiere · Sep 17, 2026131 points
- Gemma-4-26B-A4B with 39-69% less thinking at same accuracy

gemma-4-31B-it

gemma-4-26B-A4B-itarmcat · Oct 23, 2026127 points
- Shapelearn Qwen 3.8 27B (13.1 GB VRAM)

Qwen3.8-27Bsyntaxing · Sep 18, 2026104 points
- Implementation of GPT-2 in pure CMake
gpt2porridgeraisin · Aug 23, 202693 points - Due to concerns about malicious applications, GPT2 will not be released (2019)
gpt2Jimmc414 · Sep 13, 202674 points - Kimi K3.1 Pro

Kimi-K3.1-Pronekofneko · Oct 23, 202661 points
- DeepSeek v4.1 Flash Uncensored

DeepSeek-V4.1-Flashsoltanov · Sep 11, 202661 points
- Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh

Qwen3.8-27Bkisjovan · Sep 16, 202633 points
- Hyper Optimizing Kimi K3 in vLLM

Kimi-K3torutofu · Oct 18, 202631 points
- Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)

Qwen3.8-27Banemll · Oct 17, 202630 points
- Neuron Explorer for Qwen3.5-4B: See every neuron at work, make targeted edits

Qwen3.5-2B

Qwen3.5-27BSchlagbohrer · Oct 26, 202629 points
- Run QWEN3.8 27B on 16gb Nvidia GPUs

Qwen3.8-27BPragmata · Sep 17, 202628 points
- Alibaba hikes DeepSeek V4 Flash 0731 output prices by 230%

DeepSeek-V4-Flash-0731

DeepSeek-V4-FlashGloVin · Oct 21, 202626 points
- Why didn't we get GPT-2 in 2005?
gpt2bilsbie · Oct 2, 202626 points - DeepSeek V4 Pro 0813

DeepSeek-V4-Pronthypes · Nov 14, 202625 points
- Implementing GPT-2 in CSS
gpt2zapt02 · Nov 7, 202622 points - DeepSeek v4.1 flash runs 23 seconds/token on a 2020 16gb M1 Mac Mini

DeepSeek-V4.1-Flashajay-higgs · Sep 12, 202622 points
- DeepSeek v4.1 Flash is now available for internal beta testing

DeepSeek-V4.1-Flashdares2573 · Sep 8, 202622 points
- Show HN: Maple – Sub-2-BPW Moe Pruning to Run Gemma 4 26B at 45 TPS on a 6GB GPU

gemma-4-31B-it

gemma-4-26B-A4B-itAmSach · Nov 8, 202621 points