large language models

Recent window: last 5.4 months (2026-05-01 → 2026-10-11), compared with the prior 5.4 months.

horizontal · 143 members · Data as of 2026-10-14

Products in this niche provide foundational, specialized, and distilled neural network architectures trained for natural language understanding, coding, and reasoning. They are built for software developers, machine learning researchers, and enterprise engineering teams deploying intelligent applications and autonomous agents. Unlike end-user AI applications or workflow tools, these products focus directly on the underlying model weights, training methods, and core inference engines.

This theme's recent-window count is below the current min-count filter, so it does not appear in the opportunity scorecard.

Pulled from real Hacker News comment threads on this niche's launches.

Most asked for

2-bit and 4-bit quantization models (1)
Raised on 1-Bit Bonsai, the First Commercially Viable 1-Bit LLMs
Ability to decode chain of thought from latent representation (1)
Raised on DeepSeek-V4 Latent Reasoning
Achieve parity with modern 8B model performance (1)
Raised on Steerling-8B, a language model that can explain any token it generates
Add license header to source file (1)
Raised on Linear RNN/Reservoir hybrid generative model, one C file (no deps.)
API access for external integration (1)
Raised on Flint
Assess output consistency by rerunning and comparing results (1)
Raised on Cactus Hybrid: We taught Gemma 4 to know when it's wrong
Auto-complete for terminal/CLI commands (1)
Raised on Sweep, Open-weights 1.5B model for next-edit autocomplete
axis labels and explanations for graphs (1)
Raised on Mini-AGI

Concerns raised

$200 minimum monthly subscription pricing (1)
Raised on Pingu Unchained an Unrestricted LLM for High-Risk AI Security Research
annotation costs are extremely high ($50-$1000+ per annotation) (1)
Raised on Needle: We Distilled Gemini Tool Calling into a 26M Model
Battery drain on edge devices in production (1)
Raised on Cactus Hybrid: We taught Gemma 4 to know when it's wrong
calling it AGI is inaccurate and misleading (1)
Raised on Mini-AGI
Censorship in Chinese models on sensitive topics (religion, politics, Tiananmen Square) (1)
Raised on Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
Chinese models' state-approved content filtering (1)
Raised on Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
Code and writing may be AI-generated without proper human verification (1)
Raised on DeepSeek-V4 Latent Reasoning
Code duplication instead of abstraction in multi-turn sessions (1)
Raised on Sweep, Open-weights 1.5B model for next-edit autocomplete

Compared to

Anthropic's chain-of-thought methods (1)
Raised on DeepSeek-V4 Latent Reasoning
ATTRIBUTION.md protocol (1)
Raised on Steerling-8B, a language model that can explain any token it generates
autoregressive models (1)
Raised on I trained a language model that thinks the capital of Japan is Paris
char-mdlm (1)
Raised on Tiny Diffusion
Claude (1)
Raised on Pingu Unchained an Unrestricted LLM for High-Risk AI Security Research
Codeium (1)
Raised on Sweep, Open-weights 1.5B model for next-edit autocomplete
Codex (1)
Raised on I trained a language model that thinks the capital of Japan is Paris
CoLaR paper approaches (1)
Raised on DeepSeek-V4 Latent Reasoning

Members

Name	Source	Upvotes ▲	Launched
opendecider	GITHUB	13	2026-09-27
PewDiePie-Ajax-Uncensored-AI	GITHUB	14	2026-10-03
I trained a language model that thinks the capital of Japan is Paris	HN	15	2026-07-05
folia-sonnet	GITHUB	17	2026-10-09
FRIDA-Decisions	GITHUB	18	2026-10-02
Zero-power photonic language model–code	HN	18	2025-11-29
qwen38-flashnext-exl3	GITHUB	18	2026-09-17
qwen38-exl3-dflash2	GITHUB	18	2026-09-15
openjev	GITHUB	19	2026-09-16
I shrank DeepSeek V4 Flash to 57GB and it wrote a compiler on my Mac	HN	21	2026-08-16
Korean-llm-v4	GITHUB	21	2026-09-09
eikos	GITHUB	25	2026-09-23
Mica-v0.1-4B	GITHUB	26	2026-09-25
GLM-5.3-DWM32-A2.2-Recipe	GITHUB	26	2026-10-10
Confucius4-T3PO	GITHUB	27	2026-09-14
gutsy	GITHUB	28	2026-10-01
The Analog I	HN	29	2026-01-16
SHADOW-50M-Instruct	GITHUB	29	2026-09-15
DeepSeek-V4 Latent Reasoning	HN	30	2026-08-09
Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh	HN	30	2026-09-16
Hemmingway-1	GITHUB	34	2026-09-20
KaLM-Jev	GITHUB	35	2026-09-21
giga-embeddings-10B-A1.8B_hybrid	GITHUB	41	2026-09-25
openthai-systemone	GITHUB	49	2026-09-20
HC-DLM	GITHUB	53	2026-09-29
Phocinae-Largha-150M-v1	GITHUB	60	2026-10-08
flm	GITHUB	71	2026-09-15
Xing4.0-29B-A4B	GITHUB	81	2026-09-17
Grok 4.7	PH	93	2026-09-22
CUA-S1	HN	93	2026-09-19
reflex	GITHUB	101	2026-09-17
LoopVL	GITHUB	102	2026-10-01
StartLux-Decision	GITHUB	130	2026-09-29
decider	GITHUB	130	2026-09-16
Naive-N0.5-Flash	GITHUB	142	2026-09-27
tev1	GITHUB	154	2026-09-23
LongCat-2.0	PH	154	2026-07-07
YANchor	GITHUB	155	2026-10-07
Gemini 3.1 Flash-Lite	PH	168	2026-05-16
Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it	HN	170	2026-07-30
Mercury Edit 2	PH	171	2026-04-04
Grok 4.5	PH	172	2026-07-27
Tiny Diffusion	HN	172	2025-11-10
Microsoft-Decision-1	PH	174	2026-10-10
Nemotron 3 Ultra by NVIDIA	PH	179	2026-06-05
Inkling	PH	181	2026-07-20
Gemini 3.6 Flash Family	PH	182	2026-07-22
Mistral Large 4	PH	183	2026-10-07
Step 3.7 Flash	PH	183	2026-05-30
Cognition's SWE-2	PH	189	2026-09-13