Self-hosted

238 entities · 24 projects · 84 tools · 111 models · 19 news

D
Model Sep 22, 2026
DeepSeek's published DeepSWE v1.1 chart plotting pass rates against average task cost for Flash 1017 and comparison models.

DeepSeek V4.1 Flash

An MIT-licensed DeepSeek V4.1 checkpoint for fast multimodal text generation.

By
DeepSeek
Access
Not recorded
License
Not recorded
Deployment
Not recorded
Signal 512k downloads
multimodalself-hosted 2 sources
M
Model Nov 2, 2026

Moebius Inpaint GGUF

Experimental community GGUF conversions of Moebius image-inpainting weights for CrispEmbed's native engine, with four published precision variants.

By
cstr; upstream Moebius by hustvl
Access
Public GGUF downloads; requires the experimental CrispEmbed native inpainting engine and matching SDXL VAE
License
Upstream records differ between Apache-2.0 and MIT for weights; publisher preserves both without claiming dual licensing
Deployment
Not recorded
Signal source-backed
self-hosted 2 sources
Q
Model Oct 21, 2026
Turboderp's Qwen3.8-Flash-Next chart compares mean KL divergence against bits per weight for EXL3, self-calibrated EXL3, GGUF and NVFP4 configurations.

Qwen3.8-Flash-Next EXL3

Community EXL3 quantizations of Qwen3.8-Flash-Next, including a self-calibrated 1.66bpw branch whose author claims single-RTX-5090 agentic inference with a 64K quantized cache.

By
turboderp
Access
Downloadable EXL3 branches for compatible local inference; smallest branch uses mixed tensor precisions and a separately stored n-gram table.
License
Qwen Community License 1.0; commercial service and AI Work Assistant restrictions apply.
Deployment
Not recorded
Signal source-backed
self-hostedcoding-agents 3 sources
N
Model Oct 21, 2026
NVIDIA's Kumo Relational architecture diagram links users, orders and items, then applies attention within and across tables.

NVIDIA Kumo Relational

A pretrained relational foundation model for classification and regression across linked tables using historical labeled examples as inference context.

By
NVIDIA
Access
Downloadable weights through structured-data-models; GPU inference and a separate NVIDIA NIM serving route are documented.
License
OpenMDW 1.1 for weights; the NVIDIA-authored inference library code is Apache 2.0.
Deployment
Downloadable weights
Signal source-backed
decision-modelsself-hosted 3 sources
M
Model Oct 16, 2026
Hugging Face's oMLX Chat screenshot with Muse Spark Open selected and its reasoning effort control set to High.

Muse Spark Open

Meta's 816B-total, 42B-active mixture-of-experts model for coding and agentic work, with multimodal inputs and community GGUF quantizations.

By
Meta Superintelligence Labs
Access
Gated original weights; community GGUF variants are listed separately. Large-memory hardware and a compatible runtime are required.
License
Muse Model License, with usage, attribution and qualifying Model-as-a-Service revenue terms; not an unrestricted permissive license.
Deployment
Not recorded
Signal source-backed
multimodalself-hostedcoding-agents 3 sources
F
Model Oct 16, 2026

Feyospace S1 OBLITERATED weighted GGUF

Community weighted GGUF conversions of OBLITERATUS's refusal-modified 28B Feyospace cybersecurity model, with listed files from 11.0 to 22.5 GB.

By
mradermacher; parent modification by OBLITERATUS; original model by Feyospace Team at Vera Praxis
Access
Public downloadable weighted/imatrix GGUF files; runtime compatibility and memory overhead need separate checking.
License
Apache-2.0 declared by the conversion and OBLITERATUS model cards.
Deployment
Not recorded
Signal source-backed
self-hostedagent-security 2 sources
P
Model Oct 15, 2026
Daily's PhoneLLM Max Alpha 2 example conversation shows short reasoning sentences, a changed party size, a reservation tool call and an end-call tool call.

PhoneLLM Alpha 2

Pipecat's Mini and Max text models for voice-agent pipelines, with tool calling, Pipecat Flows training and short reasoning in Max.

By
Daily / Pipecat
Access
Downloadable Mini and Max weights, including quantized variants; local inference requires compatible NVIDIA hardware and serving software.
License
Mini declares BSD-2-Clause with continuing NVIDIA base-model terms; Max declares OpenMDW 1.1. Consult the individual checkpoint licenses.
Deployment
Not recorded
Signal source-backed
ai-agentsself-hosted 3 sources
F
Model Oct 15, 2026
Author-generated FLUX 3 Klein sample pairs, with BF16 on the left and MXFP8 on the right: a sunflower portrait, a dog and a Tokyo street.

FLUX 3 Klein MXFP8 by FP8Enjoyer

A community MXFP8 conversion of the distilled FLUX 3 Image Klein transformer for ComfyUI, with laptop timings, paired samples and a compatibility patch.

By
FP8Enjoyer; original model by Black Forest Labs
Access
Downloadable ComfyUI-native Safetensors diffusion transformer; matching text encoder and VAE are separate.
License
FLUX Kommunity License v1.0; derivative restrictions and conditional commercial-output terms remain applicable.
Deployment
Not recorded
Signal source-backed
multimodalself-hosted 2 sources
F
Model Oct 15, 2026
Hugging Face's generated knitted knight and dragon scene, with red boxes marking the knight's source and destination positions.

FLUX 3 Image Klein

Black Forest Labs' downloadable image-generation and editing model, with a 5.6B image transformer, multiple references and prompt-based layout boxes.

By
Black Forest Labs
Access
Downloadable base and distilled weights with gated file access; hosted BFL API and a Hugging Face demo.
License
FLUX Kommunity License v1.0; non-commercial uses and a conditional commercial-output exception below USD5 million gross annualized revenue including affiliates.
Deployment
Not recorded
Signal source-backed
multimodalself-hosted 4 sources