Tiel-Coder-35B and GLM-5.3 top Hugging Face, MiniMax H3 on Fal, and DeepSeek V4 batch (new models for daily use)
TL;DR
Hugging Face saw five models trend including Tiel-Coder-35B-A3B-GGUF and GLM-5.3 while Fal added MiniMax H3 Max video, Cohere released Parse 5, and DeepSeek V4 Flash 731 reached OpenRouter for batch coding work.
What shipped
On 29 August multiple model releases landed on public hubs and marketplaces. Hugging Face led with five trending entries that cover text, image, and video tasks. Builders now have fresh options to test without new infrastructure.
Hugging Face trending
Five models climbed the Hugging Face charts today. Tiel-Coder-35B-A3B-GGUF and Qwen3.8-Flash-Next-FP8 handle image-text work while GLM-5.3 and phonellm-alpha-1 focus on text. FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree adds a fast text-to-video path.
- •Tiel-Coder-35B-A3B-GGUF peculiar-ragdoll released Tiel-Coder-35B-A3B-GGUF, an image-text-to-text model on Hugging Face. It supports fine-tuning and inference for multimodal tasks. Vibe builders can test image and text inputs directly on the Hub.
- •GLM-5.3 zai-org released GLM-5.3, a text-generation model on Hugging Face. It runs via the transformers library for quick fine-tuning. Developers gain a ready option for general text tasks without extra setup.
- •Qwen3.8-Flash-Next-FP8 Qwen released Qwen3.8-Flash-Next-FP8, an image-text-to-text model on Hugging Face. It supports fast inference for mixed inputs. SMB owners can process image captions or simple reports in one flow.
- •FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree FastVideo released FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree, a text-to-video model on Hugging Face. It produces short clips from prompts. Creators can generate stylized video without heavy local hardware.
- •phonellm-alpha-1 pipecat-ai released phonellm-alpha-1, a text-generation model on Hugging Face. It targets phone-style conversations. Builders can prototype voice agents with low setup cost.
Fal model gallery
MiniMax H3 Max: fal released MiniMax H3 Max text-to-video tuned for stronger prompt adherence and aesthetics. It delivers higher throughput on their stack. Video creators get stylized outputs with lipsync at usable quality.
Product Hunt picks
Cohere Parse 5: Cohere released Parse 5 to turn complex documents, tables, and images into AI-ready data. SMB owners can extract invoice details or report figures faster than manual review.
Industry news
DeepSeek V4 Flash 731: DeepSeek V4 Flash 731 (batch) launched on OpenRouter with 1,049k context at $0.14 per million input tokens. It uses 13B active parameters for coding and agents. Developers can run long-context batch jobs without high spend.
Other
DeepSeek: DeepSeek V4 Flash 731 (batch) now available on OpenRouter (1,049k context, $0.14/M in, $0.28/M out): DeepSeek V4 Flash 731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows
What this means for you
For Vibe Builders: You can now test Tiel-Coder-35B-A3B-GGUF and Qwen3.8-Flash-Next-FP8 on Hugging Face for image-text work and try MiniMax H3 Max on Fal for short video clips. Cohere Parse 5 lets you turn documents into clean data without code. These options reduce the need for custom pipelines when shipping quick prototypes.
For Non-techies: For your business, new tools like Cohere Parse 5 convert invoices and reports into usable data while FastVideo and MiniMax H3 Max generate short clips from text. DeepSeek V4 on OpenRouter offers cheap long-context runs for simple agent tasks. You can start with these hosted options instead of hiring help.
For Developers: On the platform side, five Hugging Face models and DeepSeek V4 Flash 731 on OpenRouter give concrete benchmarks for image-text, video, and large-context coding. Compare GLM-5.3 and phonellm-alpha-1 against your current stack for text agents. Test prompt adherence on MiniMax H3 Max before adding it to production video flows.
What to watch next
Track whether Tiel-Coder-35B-A3B-GGUF or Qwen3.8-Flash-Next-FP8 move into top downloaded spots on Hugging Face. Watch for OpenRouter pricing changes on DeepSeek V4 Flash 731 batch. Check Fal for any follow-up MiniMax variants this week.
Harsh’s take
The day shows another wave of hosted models that lower the bar for quick tests yet still require users to pick the right fit. Most releases repeat familiar capabilities rather than open new ground. The practical move is to run one short benchmark on your own prompts before adding any model to a live workflow.
Open data talk from VZVC highlights a real gap. Many new models stay behind paywalls while shared datasets remain small. Builders who pull public model cards and run local evals this week will spot reliability issues faster than those who wait for vendor claims.
by Harsh Desai
Sources
Hugging Face trending
- •Tiel-Coder-35B-A3B-GGUF by peculiar-ragdoll trends on HuggingFace
- •GLM-5.3 by zai-org trends on HuggingFace
- •Qwen3.8-Flash-Next-FP8 by Qwen trends on HuggingFace
- •FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree by FastVideo trends on HuggingFace
- •phonellm-alpha-1 by pipecat-ai trends on HuggingFace
Fal model gallery
Product Hunt picks
Industry news
Other
More AI news
- Daily RoundupVercel eve agents and CLI updates, Hy4 on AI Gateway, and new model drops for agents
Vercel pushes agent deployment and CLI tools while new models from Tencent, unsloth, and others land on gateways and hubs for immediate testing and integration.