Entity trail

Hunyuan 3

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-07 / sources-researcherWorldClaw Generates Explorable 3D Open Worlds From Text Using a Multi-Agent Coarse-to-Fine Pipeline That Keeps Assets EditableTencent Hunyuan's WorldClaw (arXiv 2608.05248, Aug 5) uses planning agents to convert a prompt into specifications for regions, terrain, assets, materials and spatial relations, then builds the scene in stages — semantic-layout terrain, reusable asset placement, generative or procedural materials, and terrain-conditioned refinement with textured mesh reconstruction — before render-based agents polish appearance and physical interaction. The stated design goal is the one that usually gets sacrificed in text-to-3D: preserving global spatial coherence and explicit, editable assets rather than emitting an opaque neural field. No comparative benchmark numbers are reported against prior open-world generators, which is the main reason to hold judgment.
2026-08-05 / sources-researcherTencent's Hunyuan3D-Buffalo 1.0 Unifies 3D Generation, Understanding and Editing on an 87M-Scale Multimodal Corpus, Cutting Edit Chamfer Distance 86.7%Published August 5, 2026 (arXiv 2608.02711) and third on Hugging Face Daily Papers with 65 upvotes, Hunyuan3D-Buffalo bolts a Qwen-VL backbone onto a diffusion generator initialized from Hunyuan3D-2.1 to cover 3D understanding, text-to-3D, instruction-guided editing and text-grounded part generation in one model, trained on a purpose-built 87M-scale 3D multimodal corpus. Human evaluation gives it 56.6% overall preference against a 25% random baseline; on Edit3D-Bench it reports Chamfer Distance 0.0091 (an 86.7% reduction) and F1 0.6515 (2.39x the strongest baseline); UniPart-Bench part Q&A hits 85.47 SBERT. The paper does not state parameter counts, and neither weight availability nor a license is specified — worth checking before planning around it.
2026-07-15 / sources-researcherTencent Releases 1-bit and 4-bit Hunyuan 3 — a 295B Model Deployable on a Single GPUTencent published 1-bit and 4-bit compressed versions of Hunyuan 3, a 295B-parameter model, engineered to run on a single GPU. It's part of a clear July trend of aggressive quantization from Chinese labs pushing frontier-scale models onto commodity hardware. For builders, single-GPU deployment of a near-300B model changes the self-hosting math for anyone who previously assumed multi-GPU clusters were the entry ticket.
2026-07-15 / reddit-researcherllama.cpp Merges Native Hy3 Support (PR #25395) — Tencent's 295B MoE Now Runs Locally via 1-Bit/4-Bit GGUFOn July 14, 2026, llama.cpp merged upstream support for Tencent's hy_v3 (Hunyuan Hy3) architecture, a 295B-parameter / 21B-active MoE, so any recent master build can now load it; community GGUF quants (Q2_K, IQ2_M, Q4_K_M) from AngelSlim and others already ship on Hugging Face, with a 1M-context conversion reported running at ~11 tok/s. The PR also wires up MTP speculative decoding (--spec-type draft-mtp, --spec-draft-p-min 0.75), measured at roughly +40% decode throughput on code. This is the local-inference milestone of the week: a near-frontier 295B open model becomes runnable on a single high-memory rig, days after Hy3's Apache-2.0 July 6 release.
2026-07-13 / sources-researcherLong-Horizon-Terminal-Bench: Best Frontier Agent Clears Just 15% of Real Terminal Tasks, Mean 4.3%Tencent Hunyuan's new benchmark (arXiv:2607.08964, HuggingFace Daily Papers #1 today with 41 votes) puts 15 frontier models against 46 long-horizon terminal tasks across nine categories — experiment reproduction, software engineering, multimodal analysis, interactive games, and scientific computing — using dense reward-based grading that scores partial progress via graded subtasks rather than pass/fail. The strongest model reached only 15.2% pass@1 at a 0.95 partial-reward threshold (10.9% at perfect 1.0), with the cross-model mean at 4.3% and 1.7%. Tasks are brutally expensive — roughly 9.9M tokens, 231 episodes, and 85 minutes of execution each — a sobering reality check for anyone shipping autonomous CLI agents.
2026-06-14 / news-researcherChinese Open-Weight Models Now Hold 5 of Hugging Face's Top-10 Trending SlotsA mid-June Hugging Face trending snapshot shows Chinese open-weight models — including DeepSeek V4.1 Flash, Qwen, GLM, Kimi K2.7-Code, and MiniMax — occupying five of the top ten slots, the highest concentration on record, following a two-week release wave (Qwen 3.7, DeepSeek V4.1, Hunyuan Large 3, ERNIE 5.1, GLM-6). For builders, permissive licenses and low cost-per-token are driving real migration of everyday workloads to Chinese open weights.
2026-05-16 / hn-researcherImage-Blaster: Single Image to Full 3D Environment in Under 5 Minutes — 180 Points on HNOpen-source tool 'image-blaster' combines Claude AI skills with World Labs and FAL APIs to convert a single image into meshed 3D environments. Outputs include .glb/.obj meshes via Hunyuan 3D, Gaussian splats via World Labs Marble, and ambient sound effects via ElevenLabs. Dynamic objects are extracted as standalone meshes while being removed from the source image for clean environment generation. Compatible with Unity, Unreal, Godot, Blender, and Three.js — significant for game dev and spatial computing pipelines.
2026-05-01 / sources-researcherTencent Open-Sources HY-MT1.5: 440MB Offline Translation Model That Outperforms Google Translate Across 33 LanguagesTencent Hunyuan Team open-sourced HY-MT1.5-1.8B-1.25bit, a 440MB on-device translation model supporting 33 languages and 1,056 translation directions. Uses Sherry 1.25bit quantization (accepted at ACL 2026) with 3:4 fine-grained sparsity to compress from 3.3GB FP16 to 440MB with minimal accuracy loss. Outperforms Tower-Plus-72B, Qwen3-32B, and commercial APIs including Microsoft Translator. Runs fully offline on phones via custom STQ mobile CPU kernel — no subscription, no internet required.

Graph relationships

RELEASED
Tencent
Tencent released 1-bit and 4-bit compressed versions of Hunyuan 3, a 295B-parameter model.
Source finding
RELEASED
Tencent -> Hunyuan 3
Tencent released 1-bit and 4-bit compressed versions of Hunyuan 3, a 295B-parameter model.
Source finding

Source trail

arXiv / HuggingFace Daily Papers (41 upvotes)arXiv / HuggingFace Daily Papers Latent Space (AINews) / Tencent llama.cpp (GitHub PR #25395) / Hugging Face arXiv (Tencent Hunyuan) / HuggingFace Daily Papers Hugging Face Trending (Presenc AI)GitHub / Hacker News Tencent / GitHub

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…