aliteq.

Type to search every guide, tool and GPU page.

Live · from our model database

New open AI models, and what each one needs to run

Some links on this page are referral links: we may earn a commission, at no cost to you.

The newest open models we track are Inkling 2 Mini (Nov 20, 2026), Tencent Hy4 Flash (Nov 17, 2026) and Inkling 2 Small (Nov 4, 2026). Inkling 2 Mini needs about 16 GB at Q4_K_M, the smallest consumer card that runs it is a 20 GB AMD Radeon RX 7900 XT, a Mac with 32 GB of unified memory also works. Tencent Hy4 Flash needs about 125 GB at Q4_K_M, no single consumer graphics card holds it, it fits a Mac with 192 GB of unified memory. Inkling 2 Small needs about 368 GB at NVFP4, no single consumer graphics card holds it, it fits a Mac with 512 GB of unified memory. Every row below is computed with the same engine as our cost-to-run pages, so the numbers match.

63 models, newest release first (latest Nov 20, 2026). Rental prices captured , refreshed hourly. Memory at a 16K context.

Every open model, newest first

Memory = weights + context cache + about 0.8 GB for the runtime. Smallest GPU and Mac are the smallest that run the 4-bit size with 10% to spare (Macs: 8 GB left for macOS first). Rent is the card with the lowest typical price that fits: typical is the median of live offers (from a provider with 3+ offers or a list price), the cheapest listing is one offer.

Model~2-bit4-bit8-bitSmallest GPUSmallest MacRent nowRuns in
Inkling 2 Mini (MoE)Inkling · Nov 20, 202626.4B total · 3.4B active12 GBQ2_K16 GBQ4_K_M27 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiOllama yesLM Studio yesllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
Tencent Hy4 Flash (MoE)Hunyuan · Nov 17, 2026220B total · 11.3B active87 GBQ2_K125 GBQ4_K_M219 GBQ8_0No single consumer card192 GBAMD Instinct MI300X · 192 GB$2.39/hr typical$2.39/hr cheapest listingRunpodOllama not yetLM Studio not yetllama.cpp not yetMLX not yetvLLM yeschecked Nov 21, 2026
Inkling 2 Small (MoE)Inkling · Nov 4, 2026594B total · 17.0B active—368 GBNVFP4—No single consumer card512 GBNo single GPU we price; needs a multi-GPU serverOllama cloud onlyLM Studio not yetllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
MiniMax-M3.1-Flash (MoE)MiniMax · Oct 23, 2026427B total · 23.0B active—236 GBNVFP4—No single consumer card512 GBAMD Instinct MI350X · 288 GB$5.49/hr typical$5.49/hr cheapest listingRunpodOllama cloud onlyLM Studio not yetllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
NVIDIA Nemotron 3.5 Super VL 120B-A12B (MoE)Nemotron · Oct 12, 2026124B total · 12.0B active49 GBQ2_K71 GBQ4_K_M124 GBQ8_0No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiOllama yesLM Studio yesllama.cpp yesMLX not yetvLLM yeschecked Nov 21, 2026
DeepSeek V4.1 Flash (MoE)DeepSeek · Sep 10, 2026763B total · 16.0B active—476 GBFP4 + FP8—No single consumer cardOver 512 GBNo single GPU we price; needs a multi-GPU serverOllama cloud onlyLM Studio cloud onlyllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
GLM-5.3 (MoE)GLM · Aug 25, 2026753B total296 GBQ2_K427 GBQ4_K_M706 GBFP8No single consumer card512 GBNo single GPU we price; needs a multi-GPU serverOllama cloud onlyLM Studio cloud onlyllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
Qwen3.8-Flash-Next (MoE)Qwen · Aug 24, 2026180B total · 6.0B active71 GBQ2_K103 GBQ4_K_M179 GBQ8_0No single consumer card128 GBAMD Instinct MI300X · 192 GB$2.39/hr typical$2.39/hr cheapest listingRunpodOllama yesLM Studio yesllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
Ornith-1.5 35B-A3B (MoE)Ornith · Aug 18, 202636.0B total · 3.0B active15 GBQ2_K21 GBQ4_K_M37 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiOllama yesLM Studio not yetllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
NVIDIA Nemotron 3.5 Lightning 30B-A3B (MoE)Nemotron · Aug 11, 202631.6B total · 3.0B active13 GBQ2_K19 GBQ4_K_M32 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiOllama yesLM Studio yesllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
Qwen3.8 27BQwen · Aug 5, 202627.8B total13 GBQ2_K17 GBQ4_K_M29 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiOllama yesLM Studio yesllama.cpp yesMLX yesvLLM yeschecked Nov 21, 2026
Kimi K3 (MoE)Kimi · Jul 27, 20262.8T total · 104B active—1455 GBMXFP4—No single consumer cardOver 512 GBNo single GPU we price; needs a multi-GPU serverOllama cloud onlyLM Studio cloud onlyllama.cpp not yetMLX not yetvLLM yeschecked Nov 21, 2026
MiniMax-M3 (MoE)MiniMax · Jun 12, 2026427B total · 23.0B active169 GBQ2_K244 GBQ4_K_M425 GBQ8_0No single consumer card512 GBAMD Instinct MI350X · 288 GB$5.49/hr typical$5.49/hr cheapest listingRunpodNot checked yet
Gemma 4 12BGemma · May 23, 202612.0B total5.7 GBQ2_K7.8 GBQ4_K_M13 GBQ8_0RTX 308010 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
DeepSeek V4 Pro (MoE)DeepSeek · Apr 22, 20261.6T total · 49.0B active—806 GBFP4 + FP8—No single consumer cardOver 512 GBNo single GPU we price; needs a multi-GPU serverNot checked yet
Qwen3.6 27BQwen · Apr 21, 202627.8B total13 GBQ2_K17 GBQ4_K_M29 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Qwen3.6 35B-A3B (MoE)Qwen · Apr 15, 202636.0B total · 3.0B active15 GBQ2_K21 GBQ4_K_M37 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Mistral Small 4 119B (MoE)Mistral · Mar 16, 2026119B total · 6.5B active48 GBQ2_K69 GBQ4_K_M114 GBFP8No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Gemma 4 31BGemma · Mar 12, 202631.3B total14 GBQ2_K20 GBQ4_K_M33 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Gemma 4 26B-A4B (MoE)Gemma · Mar 12, 202625.8B total · 3.8B active11 GBQ2_K16 GBQ4_K_M27 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
NVIDIA Nemotron 3 Super 120B-A12B (MoE)Nemotron · Mar 10, 2026124B total · 12.0B active49 GBQ2_K71 GBQ4_K_M123 GBQ8_0No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Gemma 4 E4BGemma · Mar 2, 20268.0B total4.1 GBQ2_K5.5 GBQ4_K_M8.9 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Gemma 4 E2BGemma · Mar 2, 20265.1B total2.9 GBQ2_K3.8 GBQ4_K_M5.9 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Qwen3.5 9BQwen · Feb 27, 20269.7B total5.0 GBQ2_K6.7 GBQ4_K_M11 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Qwen3.5 122B-A10B (MoE)Qwen · Feb 24, 2026125B total · 10.0B active50 GBQ2_K72 GBQ4_K_M125 GBQ8_0No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Qwen3.5 35B-A3B (MoE)Qwen · Feb 24, 202636.0B total · 3.0B active15 GBQ2_K21 GBQ4_K_M37 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Qwen3.5 27BQwen · Feb 24, 202627.8B total13 GBQ2_K17 GBQ4_K_M29 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
MiniMax-M2.5 (MoE)MiniMax · Feb 12, 2026229B total94 GBQ2_K134 GBQ4_K_M219 GBFP8No single consumer card192 GBAMD Instinct MI300X · 192 GB$2.39/hr typical$2.39/hr cheapest listingRunpodNot checked yet
Qwen3-Coder-Next 80B-A3B (MoE)Qwen · Jan 30, 202679.7B total · 3.0B active32 GBQ2_K46 GBQ4_K_M80 GBQ8_0No single consumer card64 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
GLM-4.7-Flash (MoE)GLM · Jan 19, 202631.2B total · 3.0B active14 GBQ2_K19 GBQ4_K_M33 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Devstral Small 2 24BMistral · Dec 9, 202524.0B total13 GBQ2_K17 GBQ4_K_M27 GBFP8AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
NVIDIA Nemotron 3 Nano 30B-A3B (MoE)Nemotron · Dec 4, 202531.6B total · 3.5B active13 GBQ2_K19 GBQ4_K_M32 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Qwen3-VL 30B-A3B Instruct (MoE)Qwen · Sep 30, 202531.1B total · 3.3B active14 GBQ2_K20 GBQ4_K_M33 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Qwen3 4B Instruct 2507Qwen · Aug 5, 20254.0B total4.6 GBQ2_K5.3 GBQ4_K_M7.0 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
gpt-oss 120B (MoE)gpt-oss · Aug 4, 2025117B total · 5.1B active—63 GBMXFP4—No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
gpt-oss 20B (MoE)gpt-oss · Aug 4, 202520.9B total · 3.6B active—14 GBMXFP4—RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Qwen3 235B-A22B (MoE)Qwen · Jul 21, 2025235B total · 22.0B active95 GBQ2_K136 GBQ4_K_M236 GBQ8_0No single consumer card192 GBAMD Instinct MI300X · 192 GB$2.39/hr typical$2.39/hr cheapest listingRunpodNot checked yet
Qwen3 32BQwen · Apr 27, 202532.8B total18 GBQ2_K23 GBQ4_K_M37 GBQ8_0RTX 509032 GB48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
Qwen3 30B-A3B (MoE)Qwen · Apr 27, 202530.5B total · 3.3B active14 GBQ2_K20 GBQ4_K_M32 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Qwen3 14BQwen · Apr 27, 202514.8B total9.0 GBQ2_K12 GBQ4_K_M18 GBQ8_0RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Qwen3 8BQwen · Apr 27, 20258.2B total6.2 GBQ2_K7.7 GBQ4_K_M11 GBQ8_0RTX 308010 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Llama 4 Scout 17B-16E (MoE)Llama · Apr 2, 2025109B total · 17.0B active44 GBQ2_K63 GBQ4_K_M109 GBQ8_0No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Gemma 3 27BGemma · Mar 1, 202527.4B total13 GBQ2_K18 GBQ4_K_M29 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
Gemma 3 12BGemma · Mar 1, 202512.2B total6.5 GBQ2_K8.7 GBQ4_K_M14 GBQ8_0RTX 308010 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Mistral Small 24BMistral · Jan 28, 202523.6B total12 GBQ2_K17 GBQ4_K_M27 GBQ8_0AMD Radeon RX 7900 XT20 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet
DeepSeek R1 Distill 70BDeepSeek · Jan 20, 202570.6B total33 GBQ2_K46 GBQ4_K_M76 GBQ8_0No single consumer card64 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
DeepSeek R1 Distill 32BDeepSeek · Jan 20, 202532.8B total18 GBQ2_K23 GBQ4_K_M37 GBQ8_0RTX 509032 GB48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
DeepSeek R1 Distill 14BDeepSeek · Jan 20, 202514.8B total9.5 GBQ2_K12 GBQ4_K_M18 GBQ8_0RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Phi-4 14BPhi · Dec 11, 202414.7B total9.6 GBQ2_K12 GBQ4_K_M18 GBQ8_0RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Llama 3.3 70B InstructLlama · Nov 26, 202470.6B total33 GBQ2_K46 GBQ4_K_M76 GBQ8_0No single consumer card64 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Qwen2.5 Coder 32BQwen · Nov 6, 202432.8B total18 GBQ2_K23 GBQ4_K_M37 GBQ8_0RTX 509032 GB48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
Qwen2.5-Coder 14BQwen · Nov 6, 202414.8B total9.5 GBQ2_K12 GBQ4_K_M18 GBQ8_0RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Llama 3.2 3B InstructLlama · Sep 18, 20243.2B total3.8 GBQ2_K4.3 GBQ4_K_M5.7 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Qwen2.5 32B InstructQwen · Sep 17, 202432.8B total18 GBQ2_K23 GBQ4_K_M37 GBQ8_0RTX 509032 GB48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
Qwen2.5 7B InstructQwen · Sep 16, 20247.6B total4.6 GBQ2_K6.0 GBQ4_K_M9.2 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Llama 3.1 8B InstructLlama · Jul 18, 20248.0B total5.9 GBQ2_K7.3 GBQ4_K_M11 GBQ8_0RTX 308010 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Gemma 2 27BGemma · Jun 24, 202427.2B total17 GBQ2_K22 GBQ4_K_M33 GBQ8_0RTX 509032 GB48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
Gemma 2 9BGemma · Jun 24, 20249.2B total9.6 GBQ2_K11 GBQ4_K_M15 GBQ8_0RTX 5070 Ti16 GB24 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Mistral 7B Instruct v0.3Mistral · May 22, 20247.2B total5.6 GBQ2_K6.9 GBQ4_K_M10.0 GBQ8_0RTX 308010 GB16 GBT4 · 16 GB$0.055/hr typical$0.055/hr cheapest listingVast.aiNot checked yet
Mixtral 8x7B (MoE)Mistral · Dec 10, 202346.7B total21 GBQ2_K29 GBQ4_K_M49 GBQ8_0No single consumer card48 GBA40 · 48 GB$0.213/hr typical$0.160/hr cheapest listingVast.aiNot checked yet
GLM-4.5-Air (MoE)GLM · date not listed106B total45 GBQ2_K64 GBQ4_K_M109 GBQ8_0No single consumer card96 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Qwen3-Next 80B-A3B (MoE)Qwen · date not listed80.0B total32 GBQ2_K46 GBQ4_K_M80 GBQ8_0No single consumer card64 GBA100 SXM · 80 GB$0.656/hr typical$0.200/hr cheapest listingVast.aiNot checked yet
Qwen3-Coder 30B-A3B (MoE)Qwen · date not listed30.5B total14 GBQ2_K20 GBQ4_K_M32 GBQ8_0RTX 3090 Ti24 GB32 GBRTX 3090 · 24 GB$0.160/hr typical$0.067/hr cheapest listingVast.aiNot checked yet

We earn a commission if you buy through this link.

~2-bit is Q2_K, which we size on the safe side (real files of big models come out a little smaller). Models released at 4-bit (MXFP4, NVFP4, FP4) show their own download in the 4-bit column and nothing at 2-bit or 8-bit: a bigger copy stores the same numbers in more bytes, and how far sub-4-bit builds shrink depends on who made them. FP8 releases show their own weights at 8-bit. Hybrid-attention models (Qwen3.5 and newer, Gemma) count only their full-attention layers for the context cache. How we size memory: methodology.

What changed in the newest releases

From each official model card, and runtime support from the Ollama library, LM Studio, Hugging Face (GGUF and mlx-community builds) and vLLM, checked by hand on the date shown. "Cloud only" means the runtime lists it as hosted inference, not a download you can run at home.

Inkling · Nov 20, 2026

Inkling 2 Mini (MoE)

A looped MoE transformer. The 24 stored layers include a middle block of 12 that runs three times with shared weights, so each token gets 48 layer passes (3.4B active, about 5.4B effective). 26.4B total, 8 of 128 experts per token, hybrid local and global attention, text and image input, up to 1M context, shipped as one mixed NVFP4 and MXFP8 checkpoint of about 20 GB. The card makes no comparison with an earlier Inkling model. model card

Ollama yesLM Studio yesllama.cpp yesMLX yesvLLM yes

16 GB at Q4_K_M · checked Nov 21, 2026

Hunyuan · Nov 17, 2026

Tencent Hy4 Flash (MoE)

Compared with Hy4 the card lists three architecture changes. Hybrid attention (sparse IndexPool and 512-token sliding-window layers at 1 to 3), LatentMoE (16 of 512 routed experts work in a smaller latent space, plus one shared expert) and Engram n-gram embeddings (learned bigram and trigram tables read by hash lookup). 220B total including 64B of Engram tables, 12B active, text and image input, 1M context and a built-in DFly2 drafter for speculative decoding. On Tencent's WorkIndex it scores 42.1 against 39.7 for Hy4 preview at about a thirteenth of the cost per task. model card

Ollama not yetLM Studio not yetllama.cpp not yetMLX not yetvLLM yes

125 GB at Q4_K_M · checked Nov 21, 2026

Inkling · Nov 4, 2026

Inkling 2 Small (MoE)

A 48-layer MoE that routes each token to 8 of 384 experts, 594B total and 17B active, with hybrid local and global attention, text and image input and up to 1M context. Pretrained, midtrained and post-trained in NVFP4, which the card calls its native precision; the MXFP8 copy scores only 1 to 2 points higher on evals. The card makes no comparison with an earlier Inkling model. model card

Ollama cloud onlyLM Studio not yetllama.cpp yesMLX yesvLLM yes

368 GB at NVFP4 · checked Nov 21, 2026

Qwen · Aug 24, 2026

Qwen3.8-Flash-Next (MoE)

An experimental preview of the architecture planned for Qwen4. Gated DeltaNet is now paired with Qwen Sparse Attention (block-level sparse attention), with gated residual streams and an n-gram embedding added. 125B total and 6B active, plus 51B of n-gram embedding and a 4B MTP head. The card names no direct predecessor. model card

Ollama yesLM Studio yesllama.cpp yesMLX yesvLLM yes

103 GB at Q4_K_M · checked Nov 21, 2026

Nemotron · Aug 11, 2026

NVIDIA Nemotron 3.5 Lightning 30B-A3B (MoE)

30B total and 3B active, a Mamba-2, MoE and attention hybrid, released with DSpark, MTP and DFlash speculative decoding options and up to 1M context. In the card's own table it beats Nemotron 3 Nano on SWE-bench Verified (51.6 vs 34.1), Terminal-Bench 2.1 (24.6 vs 8.3) and AA-LCR long context (52.0 vs 32.8), and is about even on IFBench (71.9 vs 72.2). The card lists no architecture changes against Nano. model card

Ollama yesLM Studio yesllama.cpp yesMLX yesvLLM yes

19 GB at Q4_K_M · checked Nov 21, 2026

Qwen · Aug 5, 2026

Qwen3.8 27B

Built on the Qwen3.5 architecture (dense, 3 Gated DeltaNet layers per Gated Attention layer). The card reports gains in coding, research and long agent tasks and broader agent-harness support. Thinking is on by default and can be turned off per request. The card gives no point-by-point comparison with Qwen3.6 27B. model card

Ollama yesLM Studio yesllama.cpp yesMLX yesvLLM yes

17 GB at Q4_K_M · checked Nov 21, 2026

Each family, newest first

Which release replaced which: the same size class usually keeps its number (27B, 35B-A3B) across versions.

Inkling

  1. Inkling 2 Mini (MoE)Nov 20, 2026 · 16 GB · 20 GB card
  2. Inkling 2 Small (MoE)Nov 4, 2026 · 368 GB · 512 GB Mac

Hunyuan

  1. Tencent Hy4 Flash (MoE)Nov 17, 2026 · 125 GB · 192 GB Mac

MiniMax

  1. MiniMax-M3.1-Flash (MoE)Oct 23, 2026 · 236 GB · 512 GB Mac
  2. MiniMax-M3 (MoE)Jun 12, 2026 · 244 GB · 512 GB Mac
  3. MiniMax-M2.5 (MoE)Feb 12, 2026 · 134 GB · 192 GB Mac

Nemotron

  1. NVIDIA Nemotron 3.5 Super VL 120B-A12B (MoE)Oct 12, 2026 · 71 GB · 96 GB Mac
  2. NVIDIA Nemotron 3.5 Lightning 30B-A3B (MoE)Aug 11, 2026 · 19 GB · 24 GB card
  3. NVIDIA Nemotron 3 Super 120B-A12B (MoE)Mar 10, 2026 · 71 GB · 96 GB Mac
  4. NVIDIA Nemotron 3 Nano 30B-A3B (MoE)Dec 4, 2025 · 19 GB · 24 GB card

DeepSeek

  1. DeepSeek V4.1 Flash (MoE)Sep 10, 2026 · 476 GB
  2. DeepSeek V4 Pro (MoE)Apr 22, 2026 · 806 GB
  3. DeepSeek R1 Distill 70BJan 20, 2025 · 46 GB · 64 GB Mac
  4. DeepSeek R1 Distill 32BJan 20, 2025 · 23 GB · 32 GB card
  5. DeepSeek R1 Distill 14BJan 20, 2025 · 12 GB · 16 GB card

GLM

  1. GLM-5.3 (MoE)Aug 25, 2026 · 427 GB · 512 GB Mac
  2. GLM-4.7-Flash (MoE)Jan 19, 2026 · 19 GB · 24 GB card
  3. GLM-4.5-Air (MoE)date not listed · 64 GB · 96 GB Mac

Qwen

  1. Qwen3.8-Flash-Next (MoE)Aug 24, 2026 · 103 GB · 128 GB Mac
  2. Qwen3.8 27BAug 5, 2026 · 17 GB · 20 GB card
  3. Qwen3.6 27BApr 21, 2026 · 17 GB · 20 GB card
  4. Qwen3.6 35B-A3B (MoE)Apr 15, 2026 · 21 GB · 24 GB card
  5. Qwen3.5 9BFeb 27, 2026 · 6.7 GB · 10 GB card
  6. Qwen3.5 122B-A10B (MoE)Feb 24, 2026 · 72 GB · 96 GB Mac
  7. Qwen3.5 35B-A3B (MoE)Feb 24, 2026 · 21 GB · 24 GB card
  8. Qwen3.5 27BFeb 24, 2026 · 17 GB · 20 GB card
  9. Qwen3-Coder-Next 80B-A3B (MoE)Jan 30, 2026 · 46 GB · 64 GB Mac
  10. Qwen3-VL 30B-A3B Instruct (MoE)Sep 30, 2025 · 20 GB · 24 GB card
  11. Qwen3 4B Instruct 2507Aug 5, 2025 · 5.3 GB · 10 GB card
  12. Qwen3 235B-A22B (MoE)Jul 21, 2025 · 136 GB · 192 GB Mac
  13. Qwen3 32BApr 27, 2025 · 23 GB · 32 GB card
  14. Qwen3 30B-A3B (MoE)Apr 27, 2025 · 20 GB · 24 GB card
  15. Qwen3 14BApr 27, 2025 · 12 GB · 16 GB card
  16. Qwen3 8BApr 27, 2025 · 7.7 GB · 10 GB card
  17. Qwen2.5 Coder 32BNov 6, 2024 · 23 GB · 32 GB card
  18. Qwen2.5-Coder 14BNov 6, 2024 · 12 GB · 16 GB card
  19. Qwen2.5 32B InstructSep 17, 2024 · 23 GB · 32 GB card
  20. Qwen2.5 7B InstructSep 16, 2024 · 6.0 GB · 10 GB card
  21. Qwen3-Next 80B-A3B (MoE)date not listed · 46 GB · 64 GB Mac
  22. Qwen3-Coder 30B-A3B (MoE)date not listed · 20 GB · 24 GB card

Ornith

  1. Ornith-1.5 35B-A3B (MoE)Aug 18, 2026 · 21 GB · 24 GB card

Kimi

  1. Kimi K3 (MoE)Jul 27, 2026 · 1455 GB

Gemma

  1. Gemma 4 12BMay 23, 2026 · 7.8 GB · 10 GB card
  2. Gemma 4 31BMar 12, 2026 · 20 GB · 24 GB card
  3. Gemma 4 26B-A4B (MoE)Mar 12, 2026 · 16 GB · 20 GB card
  4. Gemma 4 E4BMar 2, 2026 · 5.5 GB · 10 GB card
  5. Gemma 4 E2BMar 2, 2026 · 3.8 GB · 10 GB card
  6. Gemma 3 27BMar 1, 2025 · 18 GB · 20 GB card
  7. Gemma 3 12BMar 1, 2025 · 8.7 GB · 10 GB card
  8. Gemma 2 27BJun 24, 2024 · 22 GB · 32 GB card
  9. Gemma 2 9BJun 24, 2024 · 11 GB · 16 GB card

Mistral

  1. Mistral Small 4 119B (MoE)Mar 16, 2026 · 69 GB · 96 GB Mac
  2. Devstral Small 2 24BDec 9, 2025 · 17 GB · 20 GB card
  3. Mistral Small 24BJan 28, 2025 · 17 GB · 20 GB card
  4. Mistral 7B Instruct v0.3May 22, 2024 · 6.9 GB · 10 GB card
  5. Mixtral 8x7B (MoE)Dec 10, 2023 · 29 GB · 48 GB Mac

gpt-oss

  1. gpt-oss 120B (MoE)Aug 4, 2025 · 63 GB · 96 GB Mac
  2. gpt-oss 20B (MoE)Aug 4, 2025 · 14 GB · 16 GB card

Llama

  1. Llama 4 Scout 17B-16E (MoE)Apr 2, 2025 · 63 GB · 96 GB Mac
  2. Llama 3.3 70B InstructNov 26, 2024 · 46 GB · 64 GB Mac
  3. Llama 3.2 3B InstructSep 18, 2024 · 4.3 GB · 10 GB card
  4. Llama 3.1 8B InstructJul 18, 2024 · 7.3 GB · 10 GB card

Phi

  1. Phi-4 14BDec 11, 2024 · 12 GB · 16 GB card

Questions people ask under every launch

What is the newest open AI model?

Of the open-weight models we track, the newest is Inkling 2 Mini, released Nov 20, 2026: 26.4B parameters with 3.4B active per token. It needs about 16 GB at Q4_K_M, the smallest consumer card that runs it is a 20 GB AMD Radeon RX 7900 XT, a Mac with 32 GB of unified memory also works.

Mini, Small or the big one? How do I pick a model size?

Pick the biggest model whose 4-bit size fits in about 90% of your graphics memory; the rest is left for the context and your desktop. In the newest line-up with several sizes, Inkling 2 Mini needs about 16 GB (a 20 GB card); Inkling 2 Small needs about 368 GB (a 512 GB Mac, no single consumer card). Of all the models released in the last six months, on a 24 or 32 GB card the biggest is Ornith-1.5 35B-A3B (about 21 GB), and on a Mac with up to 64 GB it is Ornith-1.5 35B-A3B (needs a 32 GB Mac). A smaller model at 4-bit usually beats a bigger one squeezed to 2-bit.

Which new models run on a 16 GB or 24 GB graphics card?

At 4-bit with a 16K context, of the models released in the last six months: none fits a 16 GB card; 24 GB runs Inkling 2 Mini, Ornith-1.5 35B-A3B, NVIDIA Nemotron 3.5 Lightning 30B-A3B, Qwen3.8 27B. Everything else needs a bigger card, a Mac with more unified memory, MoE offload to system RAM, or a rented GPU.

Why does a mixture-of-experts model need so much memory when only a few billion parameters are active?

Every expert has to sit in memory, even though each token only uses a few of them. Inkling 2 Mini has 26.4B parameters with 3.4B active, so it needs about 16 GB at Q4_K_M. The small active count is why it runs fast once it fits, not a smaller download. With llama.cpp you can keep the experts in system RAM (MoE offload), which is slower but works.

Got a card already? See every model your PC or Mac runs. Want to try one before buying? Live cloud GPU prices list every card we price, cheapest first.