Black Forest Labs/Text → ImageRuns on a laptop
Black Forest Labs ships FLUX.3 Image Klein
The latest open image model from the FLUX team handles both text-to-image generation and editing, and arrives ready to run in Diffusers.
New TodayDeepSeek
DeepSeek V4.1-Flash Adds Vision and Reasoning
The new MIT-licensed checkpoint folds image understanding and reasoning into DeepSeek's fast, mixture-of-experts Flash line.
Datacenter
Read the story17.9K downloads
- Context window
- 1.0M tokens
- Size
- 765.3 GB
- Precision
- INT8
Reflection/ReasoningMulti-GPU server
Reflection releases Beam 1.0, a 501B open MoE
The frontier AI startup's first open-weight model activates just 23B parameters per token while targeting reasoning, code, and tool use.
Mistral AI/Text / LLMDatacenter
Mistral Large 4 arrives as a trillion-param MoE
The new flagship is a sparse mixture-of-experts model with roughly 49B active parameters per token.
The feed
Latest releases
NVIDIA/Text / LLMRuns on a laptop
NVIDIA releases compact Dory 1.0 4B text model
The new Nemotron Labs model packs a 4-billion-parameter language model trained on NVIDIA's Nemotron datasets.
IBM/Any-to-Any
IBM's Granite Switch 4.1 Unifies Chat, Logic, Speech
A single endpoint now routes text conversation, decision-making, and speech synthesis across IBM's Apache-licensed Granite line.
Tencent/Text / LLMConsumer GPU
Tencent releases Hy-Memory 30B-A3B under Apache 2.0
A memory-focused mixture-of-experts model activates just 3B of its 30B parameters per token, and ships with a permissive license.
inclusionAI/Text / LLMDatacenter
inclusionAI releases Ling-3.1-flash under MIT license
A new hybrid mixture-of-experts text model arrives on Hugging Face with permissive licensing and a focus on fast inference.
Black Forest Labs/Text → ImageMajor releaseRuns on a laptop
Black Forest Labs ships FLUX.3 Image Klein
The latest open image model from the FLUX team handles both text-to-image generation and editing, and arrives ready to run in Diffusers.
NVIDIA/Vision-LanguageWorkstation GPU
NVIDIA's Nemotron 3.5 Super VL Pairs Vision With MoE Efficiency
The 120B-parameter vision-language model activates just 12B per token and ships in NVIDIA's NVFP4 format for leaner inference.
Momentum
Trending models
Benchmarks
Open LLM leaderboard
| # | Model | Avg. |
|---|---|---|
| 1 | MaziyarPanahi/calme-3.2-instruct-78b | 52.1 |
| 2 | MaziyarPanahi/calme-3.1-instruct-78b | 51.3 |
| 3 | dfurman/CalmeRys-78B-Orpo-v0.1 | 51.2 |
| 4 | MaziyarPanahi/calme-2.4-rys-78b | 50.8 |
| 5 | huihui-ai/Qwen2.5-72B-Instruct-abliterated | 48.1 |
| 6 | Qwen/Qwen2.5-72B-Instruct Qwen · Alibaba | 48.0 |
By type
Newest in each category
Text → Image →
Black Forest Labs/Text → ImageRuns on a laptop
Black Forest Labs ships FLUX.3 Image Klein
The latest open image model from the FLUX team handles both text-to-image generation and editing, and arrives ready to run in Diffusers.
Code →
Reflection/ReasoningMulti-GPU server
Reflection releases Beam 1.0, a 501B open MoE
The frontier AI startup's first open-weight model activates just 23B parameters per token while targeting reasoning, code, and tool use.
Text → Video →
Unknown/Text → VideoConsumer GPU
Kandinsky 6.0 Generates Video and Audio Together
Kandinsky Lab's new foundation models produce synchronized sound and motion from a text prompt, in 3B and 29B sizes.
Vision-Language →
DeepSeek/Vision-LanguageDatacenter
DeepSeek V4.1-Flash Adds Vision and Reasoning
The new MIT-licensed checkpoint folds image understanding and reasoning into DeepSeek's fast, mixture-of-experts Flash line.
On the horizon
Upcoming launches
AnnouncedAug 26, 2026
Qwen Teases 3.8-Flash-Next, a 125B Sparse MoE
Alibaba's next Qwen release pairs a large parameter pool with a tiny active footprint, promising speed without the full compute bill.
Qwen · Alibaba·Text / LLM
AnnouncedSep 9, 2026
DeepSeek Announces V4.1 Flash, a Cheaper Reasoning Model
The new mixture-of-experts model is billed as more capable than V4 Pro while costing less to run, and ships under an MIT license.
DeepSeek·Text / LLM
Announced
Zing-0.5 aims for playable worlds in real time
A 5B-parameter autoregressive world model blends action and text control to generate interactive environments on the fly.
Unknown·Any-to-Any