BLK · MODEL RELEASE TRACKER

What just dropped.

Newest local AI models, date-sorted. Every row carries quick fit verdicts for the four VRAM classes operators ask about — so you know in one glance whether to bother downloading a model before it starts loading. 60 models indexed.

For broader ecosystem news see /pulse. For the recommendation engine see /choose-my-gpu.

60 models shown · newest first

AddedModelParams8GB16GB24GB48GB96GB+
1mo agoQwen3.8 27B
qwen
27B✗△~✓✓
1mo agoCommand R7B (12-2024)
command-r
8B~✓✓✓✓
1mo agoOpenELM 3B Instruct
other
3B✓✓✓✓✓
1mo agoSmolVLM Instruct
other
2.25B✓✓✓✓✓
1mo agoDeepSeek V2 Lite Chat
deepseek
15.7B✗✓✓✓✓
1mo agoOLMo 2 1B Instruct
olmo
1.48B✓✓✓✓✓
1mo agoFalcon 3 3B Instruct
falcon
3.23B✓✓✓✓✓
1mo agoGranite 3.1 2B Instruct
granite
2.53B✓✓✓✓✓
1mo agoQwen2-VL 2B Instruct
qwen
2B✓✓✓✓✓
1mo agoTinyLlama 1.1B Chat v1.0
llama
1.1B✓✓✓✓✓
1mo agoSmolLM2 360M Instruct
other
360M✓✓✓✓✓
1mo agoSmolLM2 135M Instruct
other
135M✓✓✓✓✓
1mo agoGemma 2 2B Instruct
gemma
2.61B✓✓✓✓✓
1mo agoGemma 3 270M
gemma
270M✓✓✓✓✓
1mo agoQwen 3 1.7B
qwen
1.7B✓✓✓✓✓
1mo agoQwen 3 0.6B
qwen
600M✓✓✓✓✓
1mo agoparaphrase-multilingual-MiniLM-L12-v2
other
118M✓✓✓✓✓
1mo agoall-mpnet-base-v2
other
109M✓✓✓✓✓
1mo agoall-MiniLM-L6-v2
other
22M✓✓✓✓✓
1mo agoGOT-OCR 2.0
stepfun
580M✓✓✓✓✓
1mo agoFlorence-2 Large
other
770M✓✓✓✓✓
1mo agoColPali v1.3
gemma
3B✓✓✓✓✓
1mo agoSigLIP SO400M (patch14-384)
other
878M✓✓✓✓✓
1mo agoStable Diffusion 3.5 Medium
other
2.5B✓✓✓✓✓
1mo agoSDXL Turbo
other
2.6B✓✓✓✓✓
1mo agoFLUX.1 [schnell]
other
12B△✓✓✓✓
1mo agoFLUX.1 [dev]
other
12B△✓✓✓✓
1mo agomxbai-rerank-large-v2
other
1.54B✓✓✓✓✓
1mo agoJina Reranker v2 Base Multilingual
other
278M✓✓✓✓✓
1mo agoMultilingual E5 Large Instruct
other
560M✓✓✓✓✓
1mo agoE5 Mistral 7B Instruct
other
7.11B~✓✓✓✓
1mo agoGTE ModernBERT Base
other
149M✓✓✓✓✓
1mo agoSnowflake Arctic Embed L v2.0
other
568M✓✓✓✓✓
1mo agoJina Embeddings v3
other
572M✓✓✓✓✓
1mo agomxbai-embed-large-v1
other
335M✓✓✓✓✓
1mo agoBGE Large EN v1.5
other
335M✓✓✓✓✓
1mo agoNomic Embed Text v1.5
other
137M✓✓✓✓✓
1mo agoPiper
other
25M✓✓✓✓✓
1mo agoOrpheus 3B 0.1 FT
other
3.78B✓✓✓✓✓
1mo agoF5-TTS
other
336M✓✓✓✓✓
1mo agoXTTS v2
other
460M✓✓✓✓✓
1mo agoKokoro 82M
other
82M✓✓✓✓✓
1mo agoParakeet TDT 0.6B v2
other
600M✓✓✓✓✓
1mo agoDistil-Whisper Large v3
other
756M✓✓✓✓✓
1mo agoWhisper Small
other
244M✓✓✓✓✓
1mo agoWhisper Base
other
74M✓✓✓✓✓
1mo agoWhisper Tiny
other
39M✓✓✓✓✓
1mo agoMalhajar Mistral 7B Turkish
mistral
7.2B~✓✓✓✓
1mo agoRefinedNeuro RN TR R2
llama
8B~✓✓✓✓
1mo agoRefinedNeuro RN TR R1
llama
8B~✓✓✓✓
1mo agoYTU Turkish Gemma 9B v0.1
gemma
9.2B~✓✓✓✓
1mo agoTurkcell LLM 7B v1
mistral
7.4B~✓✓✓✓
1mo agoMistral Turkish v2 (brooqs)
mistral
7.2B~✓✓✓✓
1mo agoTrendyol LLM Asure 12B
gemma
11.8B△✓✓✓✓
1mo agoCosmos Llama 3 8B Turkish
llama
8B~✓✓✓✓
1mo agoTrendyol LLM 7B Base v0.1
llama
7B✓✓✓✓✓
1mo agoTurkish GPT-2 Large
other
770M✓✓✓✓✓
1mo agoVBART Large (Turkish Summarization)
other
400M✓✓✓✓✓
1mo agoOmni 31B Turkish Reasoning
other
31B✗△~✓✓
1mo agoGemma 4 Turkish 26B (4B active)
gemma
20.4B✗~✓✓✓
✓
Comfortable
Q4 fits with KV headroom
~
Tight
Q4 fits, small context
△
Marginal
IQ3 only, expect degradation
✗
Doesn't fit
Won't run usefully
HOW FIT VERDICTS ARE DERIVED

Quick footprint estimate at Q4_K_M: params × 0.6 GB + 1.5 GB runtime overhead. Comfortable means the rig has ≥1.4× headroom for KV cache and multi-turn context. Tight means it fits but you'll bump the ceiling on long context. Marginal means only aggressive (IQ3 / IQ2) quants work, with quality degradation. Doesn't fit means weights alone won't load without RAM offload, which crushes tok/s. Frontier-class models (400B+) render as no-fit on every single-rig VRAM class — they need multi-GPU or cloud.