AI · alternatives · 2026
24 alternatives to UltraCompress – first mathematically lossless 5-bit LLM compression
Verified AI infrastructure for regulated deployment. UltraCompress (our wedge): near-lossless 5-bit compression with SHA-256-reproducible reconstruction - prove the model in production is the one you validated. 23 architectures (0.6B-405B), Hermes-3-405B @ 1.0066x. OpenAI-compatible API. pip install
Below are 24 products that do a similar job, ranked by how close each is in meaning and then by launch-day votes.
- 1
Qwen3.6-35B-A3B▲124The open sparse MoE model for agentic coding
Apr 2026 · qwen.ai · its alternatives →
- 2
I spent a lot of time and money on this rather big side project of mine that attempts to replicate the mechanistic interpretability research on proprietary LLMs that was quite popular this year and produced great research papers by Anthropic [1], OpenAI [2] and Deepmind [3]. I am quite proud of this project and since I consider myself the target audience for HackerNews did I think that maybe some of you would appreciate this open research replication as well. Happy to answer any questions or face any feedback. Cheers [1]…
2024 · github.com · its alternatives →
- 3
TurboQuant▲295New LLM compression algorithm by Google
Mar 2026 · research.google · its alternatives →
- 4
PrismML builds 1-bit and ternary Bonsai models that pack advanced reasoning, tool use, and multimodal AI into phones, laptops, and efficient data centers.
Mar 2026 · prismml.com · its alternatives →
- 5AT
A 3.16M-parameter INT4 transformer running entirely in the on-chip memory of a Xilinx Kria KV260. Zero DRAM in the token loop, 59,965 tok/s on the fabric, bit-exact. Chat with it live.
Aug 2026 · mikeayles.com · its alternatives →
- 6
Cool trick to store 4 5-bit values in one 16-bit value. - jaamesd/BucketCompressionTrick
2018 · github.com · its alternatives →
- 7
Build your own high performance LLM inference engine in C++ and CUDA - a smaller version of vLLM - jmaczan/tiny-vllm
May 2026 · github.com · its alternatives →
- 8
Qwen3.6-27B▲120The sweet-spot open dense model for coding agents
Apr 2026 · qwen.ai · its alternatives →
- 9OA
I built an experiment that uses an overfitted transformer and arithmetic coding to compress individual files. Instead of training the model to generalize, I train a 900KB transformer to memorize a single file and predict the next byte. Those predictions are fed into an arithmetic coder to produce the compressed output. On a 100MB NYC taxi CSV, it compresses to about 7MB (~0.5 bits/byte). On a 100MB slice of enwik9, it compresses to about 21MB (~1.68 bits/byte). It's pretty slow right now (roughly 20–30 minutes of training and 45 minutes each for compression and decompression on my…
Jun 2026 · its alternatives →
- 10
Make your videos smaller with zero loss of quality
2022 · play.google.com · its alternatives →
- 11
- 12
Metal-first MoE inference for Apple Silicon with bounded SSD expert streaming and local OpenAI Chat, Responses, and Anthropic Messages endpoints. - hebrus-labs/hebrus
Jul 2026 · github.com · its alternatives →
- 13
Embeddable minimal in-memory GZIP decompression library for embedded systems - wkoszek/mini_gzip
2015 · github.com · its alternatives →
- 14
The model uses a 1024-dimensional complex Hilbert space with 32 layers of programmable Mach–Zehnder meshes (Reck architecture) and derives token probabilities directly via the Born rule. Despite using only unitary operations and no attention mechanism, a 1024×32 model achieves coherent TinyStories generation after < 1.8 hours of training on a single consumer GPU. This is Part 1 - the next step is physical implementation with $50 of optics from AliExpress.
Nov 2025 · zenodo.org · its alternatives →
- 15
Abstract page for arXiv paper 2607.23806: A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
Jul 2026 · arxiv.org · its alternatives →
- 16
Good morning HN! For a while now I have been toying with this idea and now finally have a working prototype. This project allows you to encode secret messages into ordinary looking text by using arithmetic coding with a probability model derived from an LLM. By encrypting the message and then decompressing the encrypted message using the arithmetic coder, you get output which looks just like randomly sampled output from the LLM. Except, it actually encodes your secret messages in the specific choices of tokens. By using authenticated encryption, only a user who knows the key can know that a…
2025 · github.com · its alternatives →
- 17
The Unified Latent-State Memory Fabric (UL-SMF) is a hardware-software co-designed memory compression fabric that solves the memory bottleneck in long-context Transformer inference. By combining FSQ with dynamic 16-dimensional latent mapping, UL-SMF compresses Key-Value (KV) cache tensors by up to 3
Aug 2026 · github.com · its alternatives →
- 18
HAI 1.0: 1.2GB local AI. Beats GPT-5 at coding
Jul 2026 · github.com · its alternatives →
- 19
Properties of a Novel Binary Representation of Integers using Base-𝜙 - irradix/paper.md at main · DOSAYGO-STUDIO/irradix
2024 · github.com · its alternatives →
- 20
Please forgive the sign up. It means we can give away and keep track of free inference usage/no API keys needed. Happy to send sign in credentials on request if you prefer.
2025 · enginelabs.ai · its alternatives →
- 21
A simple re-implementation of Google's sparsehash as a learning excercise. - qpfiffer/Simple-Sparsehash
2015 · github.com · its alternatives →
- 22
Minimal, readable LLM post-training experiments on one 8GB GPU. Measures forgetting, seed variance, and RL emergence. - pochenai/nano-llm-posttraining
Aug 2026 · github.com · its alternatives →
- 23Q3
Ubicloud is an open source alternative to AWS. Today, we launched our inference APIs, built with open source AI models. QwQ-32B-Preview is one of those models; and it can provide o1-like reasoning at 1% the cost. QwQ is licensed under Apache 2.0 [1] and Ubicloud under AGPL v3. We deploy open models on a cloud stack that can run anywhere. This allows us to offer great price / performance. From an accuracy standpoint, QwQ does well in math and coding domains. For example, in the MMLU-Pro Computer Science LLM Benchmark, the accuracy rankings are as follows. Claude-3.5 Sonnet (82.5),…
2025 · its alternatives →
- 24
Hi HN, I built OpenGraviton, an open-source AI inference engine that pushes the limits of running extremely large LLMs on consumer hardware. By combining 1.58-bit ternary quantization, dynamic sparsity with Top-K pruning and MoE routing, and mmap-based layer streaming, OpenGraviton can run models far larger than your system RAM—even on a Mac Mini. Early benchmarks: TinyLlama-1.1B drops from ~2GB (FP16) to ~0.24GB with ternary quantization. At 140B scale, models that normally require ~280GB fit within ~35GB packed. Optimized for Apple Silicon with Metal + C++ tensor unpacking, plus…
Mar 2026 · github.com · its alternatives →
Also compare
- Qwen3.6-35B-A3B alternatives
- Llama 3.2 Interpretability with Sparse Autoencoders alternatives
- TurboQuant alternatives
- 1-Bit Bonsai, the First Commercially Viable 1-Bit LLMs alternatives
- A tiny LLM running at 21,000 tok/s on a $250 FPGA (Live Demo) alternatives
- How to store a set of four 5-bit values in one 16-bit value alternatives
Ranked by how close each launch is in meaning, then by votes. Prices were read from each product’s own site when checked and can change. Refine with your own description →