Qwen3.6-27B and Bonsai trend on Hugging Face, BaseRT speeds Apple M5 runs
TL;DR
Two 27B models climbed Hugging Face trends for image-text and text tasks while BaseRT delivered faster local inference on Apple M5 hardware than llama.cpp or MLX.
What shipped
On 19 July two large models rose on the Hugging Face Hub. A runtime optimized for Apple silicon also surfaced on Product Hunt. The releases point to continued demand for runnable local models and faster on-device execution.
Hugging Face trending
Two models climbed the trending charts on Hugging Face today. One handles image and text inputs while the other focuses on text generation. Both provide GGUF formats for easier local use.
- •Qwen3.6-27B-Fable-Fusion DavidAU released a 27B image-text-to-text model that rose on Hugging Face trends. It supports download and fine-tuning for vision-language projects. Builders gain a new uncensored option for multimodal work.
- •Bonsai-27B-gguf prism-ml released a 27B text-generation model built on llama.cpp that is trending. Users can run it locally via the Hub. It gives an alternative for efficient text tasks on consumer hardware.
Product Hunt picks
BaseRT: BaseRT runs 6.4 times faster than llama.cpp on Apple M5 devices. Local AI apps can achieve quicker responses than with MLX. On-device developers now have a stronger option for production inference.
What this means for you
For Vibe Builders: You can download the trending Qwen and Bonsai models from Hugging Face and test them locally without writing code. BaseRT then lets you run those models faster on Apple M5 hardware. Start with a small fine-tune on one model and measure the speed gain in your next prototype.
For Non-techies: SMB owners gain simpler paths to run capable models on their own Macs instead of paying cloud fees. BaseRT cuts wait times for local tasks such as document summaries or customer replies. Test the new models through easy Hub downloads to see if they fit daily workflows.
For Developers: Engineers should benchmark BaseRT against current llama.cpp and MLX setups on M5 hardware before wider rollout. The two new 27B GGUF models offer fresh candidates for local text and vision pipelines. Run quick latency and quality checks this week to decide on integration.
What to watch next
Watch for more GGUF releases on Hugging Face and additional Apple-silicon runtimes on Product Hunt. Track speed benchmarks that compare BaseRT to MLX on production workloads.
Harsh’s take
The day shows a clear pattern: smaller teams keep shipping runnable 27B variants while one runtime targets the narrow but growing Apple M5 niche. Most users will still hit the same practical limits on memory and context length that larger labs already solved months ago. The real test is whether these releases move beyond trend lists into daily local pipelines. Builders should install BaseRT on an M5 machine and run one of the new models against their current stack before the next wave of similar tools arrives.
by Harsh Desai
Sources
Hugging Face trending
- •Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF by DavidAU trends on HuggingFace
- •Bonsai-27B-gguf by prism-ml trends on HuggingFace
Product Hunt picks
More AI news
- Model ReleaseMeituan LongCat 2.0 added on OpenRouter
Meituan LongCat 2.0 is now available on OpenRouter with 1,049K context at $0.30/M input and $1.20/M output.