Skip to content
AI Compute Radar
IndexCollected 3 h ago
Field guide — Issue 071Tuesday · 10 Nov 2026 · Models + compute decision layer

Four questions.Four answers.

We read model files, model configs and public GPU rental pages, and turn them into one sentence each. Every sentence says when we read it. The numbers behind it are one click away.

One decision, not 50 tabs.

What should I use?

Say the job. Get three.

Chat, code, long documents, images, video or training: at most three open models for your card, from a list kept by a published rule. No scores.

Lists checked every Monday · ordered for your card

What does a GPU hour cost today?

RTX 4090 from $0.34

$0.50 typical on the Vast marketplace

Up from last week. Eight classes, eleven sources, every price read from the public page. Nothing is blended.

Collected 8 min ago · RunPod list price · Vast middle of 64 offers

  • ▼Qwen3.8 27B · HF trending 320
  • ▲LFM3 1.7B · HF trending 217
  • ▼Qwen3.8 Flash-Next · HF trending 123
  • ▲GLM-5.3-Flash · HF trending 59
  • ▲RTX 3090 $0.25/h
  • ◆RTX 4090 $0.50/h
  • ◆L40S $0.80/h
  • ▲RTX 5090 $0.80/h
  • ▼A100 $0.85/h
  • ▼H100 $2.92/h
  • ▲H200 $5.66/h
  • ▼B200 $7.50/h

Today’s edition — Issue 071 · Tuesday · 10 Nov 2026

Since the previous edition · newest reading 2026-11-10 15:08 UTC

  1. More popular+5KAT-Coder V2.5 · 23 → 28 · popularity2026-11-09 18:15 UTC → 2026-11-10 15:08 UTC
  2. GPU rent down−7.3%H200 · $5.68 → $5.26 per hour · Vast.ai daily median2026-11-08 23:00 UTC → 2026-11-09 23:00 UTC
  3. Trending up+59LFM3 1.7B · 158 → 217 · Hugging Face trending2026-11-09 18:15 UTC → 2026-11-10 12:15 UTC
  4. More popular+4Ornith 1.5 9B · 58 → 62 · popularity2026-11-09 18:15 UTC → 2026-11-10 15:08 UTC
  5. Fading−5GLM-5.2 · 67 → 62 · popularity2026-11-09 18:15 UTC → 2026-11-10 15:08 UTC

Pick of the week — Week 45, 2026

Granite 4.2 3B

Granite 4.2 3B by IBM takes the ninth pick: the largest popularity gain between the readings of 31 October and 6 November among models that run comfortably on a 24 GB card and are not on cooldown. Its Q4_K_M file measures 2.1 GB, read from IBM's own GGUF repository. At 8K context it needs 3.6 GB and runs easily on every card and Mac we track, the 8 GB cards included; at 32K it needs 6.3 GB, runs on the 8 GB cards and runs easily from 11 GB up.

Read the report →

00 / Radar view

Every tracked model on one scope.

A contact appears at the center when its Hugging Face repository is created and drifts outward as it ages. Brightness is its popularity. Bearing groups models by maker and means nothing else.

Radar scope of every tracked model49 models plotted by age since their Hugging Face repository was created; brightness follows popularity.30 d6 mo1 y2 yQwen3 0.6B · Alibaba Qwen · Heat 62 · repo created 2025-04-27 (562 d ago)Qwen3 14B · Alibaba Qwen · Heat 46 · repo created 2025-04-27 (562 d ago)Qwen3 32B · Alibaba Qwen · Heat 45 · repo created 2025-04-27 (562 d ago)Qwen3 8B · Alibaba Qwen · Heat 63 · repo created 2025-04-27 (562 d ago)Qwen3-Coder 30B-A3B · Alibaba Qwen · Heat 59 · repo created 2025-07-31 (467 d ago)Qwen3 4B · Alibaba Qwen · Heat 57 · repo created 2025-08-05 (462 d ago)Qwen3-Coder Next · Alibaba Qwen · Heat 45 · repo created 2026-01-30 (284 d ago)Qwen AgentWorld 35B-A3B · Alibaba Qwen · Heat 50 · repo created 2026-06-22 (141 d ago)Qwen3.8 27B · Alibaba Qwen · Heat 79 · repo created 2026-08-05 (97 d ago)Qwen3.8 27BQwen3.8 2.4T-A95B · Alibaba Qwen · Heat 54 · repo created 2026-08-08 (94 d ago)Qwen3.8 Flash-Next · Alibaba Qwen · Heat 70 · repo created 2026-08-24 (78 d ago)Qwen3.8 Flash-NextDeepSeek-R1 · DeepSeek · Heat 60 · repo created 2025-01-20 (659 d ago)DeepSeek-V4-Flash · DeepSeek · Heat 51 · repo created 2026-07-31 (102 d ago)DeepSeek-V4-Pro · DeepSeek · Heat 38 · repo created 2026-08-13 (89 d ago)Gemma 3 27B · Google · Heat 41 · repo created 2025-03-01 (619 d ago)Gemma 4 E4B · Google · Heat 66 · repo created 2026-03-02 (253 d ago)Gemma 4 26B-A4B · Google · Heat 62 · repo created 2026-03-11 (244 d ago)Gemma 4 31B · Google · Heat 61 · repo created 2026-03-11 (244 d ago)Gemma 4 12B · Google · Heat 60 · repo created 2026-05-23 (171 d ago)Granite 4.2 30B · IBM · Heat 40 · repo created 2026-08-07 (95 d ago)Granite 4.2 3B · IBM · Heat 46 · repo created 2026-08-07 (95 d ago)Granite 4.2 8B · IBM · Heat 27 · repo created 2026-08-07 (95 d ago)Ling 3.0 Flash · InclusionAI · Heat 22 · repo created 2026-08-02 (100 d ago)Ling 3.0 Tiny · InclusionAI · Heat 43 · repo created 2026-08-10 (92 d ago)KAT-Coder V2.5 · Kwaipilot · Heat 28 · repo created 2026-07-23 (110 d ago)LFM2.5 1.2B · Liquid AI · Heat 31 · repo created 2026-01-06 (308 d ago)LFM2.5 8B-A1B · Liquid AI · Heat 20 · repo created 2026-05-28 (166 d ago)LFM2.5 2.6B · Liquid AI · Heat 22 · repo created 2026-07-28 (105 d ago)LFM2.5 24B-A2B · Liquid AI · Heat 60 · repo created 2026-10-12 (29 d ago)LFM3 1.7B · Liquid AI · Heat 70 · repo created 2026-10-23 (18 d ago)Llama 3.1 8B · Meta · Heat 71 · repo created 2024-07-18 (845 d ago)Llama 3.1 8BLlama 3.2 3B · Meta · Heat 53 · repo created 2024-09-18 (783 d ago)Llama 3.3 70B · Meta · Heat 55 · repo created 2024-11-26 (714 d ago)Phi-4 Mini · Microsoft · Heat 40 · repo created 2025-02-19 (629 d ago)Mistral Small 3.2 · Mistral AI · Heat 21 · repo created 2025-06-19 (509 d ago)Devstral Small 2 24B · Mistral AI · Heat 18 · repo created 2025-11-28 (347 d ago)Nemotron Nano 9B v2 · NVIDIA · Heat 19 · repo created 2025-08-12 (455 d ago)gpt-oss 120B · OpenAI · Heat 56 · repo created 2025-08-04 (463 d ago)gpt-oss 20B · OpenAI · Heat 73 · repo created 2025-08-04 (463 d ago)gpt-oss 20BOrnith 1.5 35B-A3B · Ornith AI · Heat 55 · repo created 2026-08-18 (84 d ago)Ornith 1.5 397B · Ornith AI · Heat 42 · repo created 2026-08-18 (84 d ago)Ornith 1.5 9B · Ornith AI · Heat 62 · repo created 2026-08-18 (84 d ago)Ornith 1.5 122B · Ornith AI · Heat 56 · repo created 2026-10-16 (25 d ago)Hy3 · Tencent · Heat 36 · repo created 2026-07-02 (131 d ago)GLM-4.5-Air · Zhipu AI · Heat 38 · repo created 2025-07-20 (478 d ago)GLM-4.7-Flash · Zhipu AI · Heat 66 · repo created 2026-01-19 (295 d ago)GLM-5.2 · Zhipu AI · Heat 62 · repo created 2026-06-16 (147 d ago)GLM-5.3 · Zhipu AI · Heat 50 · repo created 2026-08-25 (77 d ago)GLM-5.3-Flash · Zhipu AI · Heat 84 · repo created 2026-08-25 (77 d ago)GLM-5.3-Flash

How to read the scope

  • Rings, from the center: 30 days, 6 months, 1 year, 2 years since the repository was created. The rim is three years and beyond.
  • Brightness: popularity, 0 to 100. A dim outline is a model still collecting its first week of history.
  • Pulsing ring: repository created within the last 30 days.
  • Bearing: maker sector, nothing more. Blips open the model page.

Newest contacts

  1. LFM3 1.7BLiquid AI · repo created 2026-10-23 · 18 d agoHeat 70
  2. Ornith 1.5 122BOrnith AI · repo created 2026-10-16 · 25 d agoHeat 56
  3. LFM2.5 24B-A2BLiquid AI · repo created 2026-10-12 · 29 d agoHeat 60
  4. GLM-5.3Zhipu AI · repo created 2026-08-25 · 77 d agoHeat 50
  5. GLM-5.3-FlashZhipu AI · repo created 2026-08-25 · 77 d agoHeat 84

Strongest returns

  1. GLM-5.3-FlashZhipu AI · 77 d ago84
  2. Qwen3.8 27BAlibaba Qwen · 97 d ago79
  3. gpt-oss 20BOpenAI · 463 d ago73

Ages come from Hugging Face repository creation dates; popularity from measured signals. How popularity is scored

01 / Model market

What is winning

Models ranked by popularity, with the measured signals behind it
RankModelHeat0–100 · compositeDownloadsHugging Face
1GLM-5.3-FlashZhipu AI · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 59$/M in: $0.156.4M+1.2% up
2Qwen3.8 27BAlibaba Qwen · OpenGeneral · Coding · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 320$/M in: $0.157.4M-7.1% down
3gpt-oss 20BOpenAI · OpenGeneral · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 21$/M in: $0.035.8M-3.1% down
4Llama 3.1 8BMeta · OpenGeneral · CodingHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 454.5M+1.3% up
5LFM3 1.7BBreakoutLiquid AI · OpenGeneralHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 217$/M in: $0.0883K+73.0% up
6Qwen3.8 Flash-NextAlibaba Qwen · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 123$/M in: $0.153.4M-11.9% down
7Gemma 4 E4BFadingGoogle · OpenGeneralHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 134.3M+2.8% up
8GLM-4.7-FlashZhipu AI · OpenCoding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 4$/M in: $0.061.1M+12.1% up
9Qwen3 8BAlibaba Qwen · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 126.8M-10.3% down
10Gemma 4 26B-A4BGoogle · OpenGeneral · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 15$/M in: $0.085.4M-7.3% down
11Qwen3 0.6BAlibaba Qwen · OpenGeneralHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 1220.5M-10.3% down
12GLM-5.2Zhipu AI · OpenCoding · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 8$/M in: $0.01319K+0.1% up

Top 12 of 49 tracked models, ranked by popularity; models still collecting history follow, ranked by trending.

Heat is a composite of measured signals only: each component is the model's percentile among models with a full week of history — trending level (35%), 7-day download growth (30%), 7-day trending change (15%), 30-day downloads (15%) and Hub likes (5%). Missing components renormalize the weights and lower the shown confidence; nothing is guessed. Trending and downloads come from the Hugging Face Hub — the download counter is a rolling 30-day window, not unique users — and input pricing from OpenRouter (CC BY 4.0). The full formula, thresholds and flag rules are on the methodology page. How Heat is computed → Every tracked model has its own page → Who leads, by maker →

02 / Compute ladder

Own, Spark or rent?

Models need machines. The right machine depends on how often you push past your own hardware — not on the biggest number on a spec sheet.

Path A — Existing PC

Use what you own.

Most people underestimate their current GPU. A 12–32 GB card runs excellent quantized models every day, free of charge.

  • + 12–32 GB VRAM class
  • + Ollama / LM Studio ready
  • + Zero extra cost

Path B — Personal AI box

Own the middle.

DGX Spark and GB10-class systems put 128 GB of unified memory on your desk. Rational when large local models are your daily routine.

  • + 128 GB unified memory
  • + 273 GB/s — capacity, not speed
  • + Private by default

Path C — Elastic compute

Rent the spike.

Occasional heavy job? Rent an H100 for the afternoon instead of buying hardware that idles the rest of the year.

  • + H100 / A100 / B200 class
  • + Pay per hour
  • + No commitment
Median $/h
  • RTX 3090$0.25up
  • RTX 4090$0.50about the same
  • L40S$0.80about the same
  • RTX 5090$0.80up
  • A100$0.85down
  • H100$2.92down
  • H200$5.66up
  • B200$7.50down

03 / My setup

Your machine, your answer

◇ Estimated— Estimated: computed from our curated model and hardware catalog — not a live reading.

Four questions. One recommendation. No benchmark knowledge required.

Runs locally in your browser — nothing is sent to any server.
Saved only in this browser (localStorage), never sent to us. Untick to forget.

Local first. — Fit score 73

Local first.

Main model
GLM-4.7-Flash · 31.2B MoE · Q4_K_M
Compute
RTX 4090 · 24 GB VRAM
Burst option
Vast.ai when the job outgrows you
Estimated memory
~18.9 GB

Why this route

  • + Weights ~17.1 GB (Q4_K_M), context ~0.4 GB at 8K tokens, runtime ~1.4 GB.
  • + Owning the inference beats paying per token at this frequency

Hugging Face ↗ (External link)

Estimated from quantized model file sizes plus runtime overhead and a safety margin. Real usage varies with context length.

The Tower Report

Every Monday, one report

Last Monday · Week 45, 2026

Granite 4.2 3B is the pick; Qwen AgentWorld 35B-A3B gained the most

The pick, who gained and who lost, what an hour cost all week, and the card you chose. Written by people; every number with the time we read it.

—

until the next Tower Report · Monday 06:00 UTC

The site reads prices every hour and popularity twice a day. The report is the one moment a week where it all stands still.

Tap the bear to wake it

From the channel

Short answers, on YouTube

The same numbers as on this site in videos under a minute: does this model fit that card, and why. Measured file sizes, computed context memory, no speed claims.

04 / Hardware radar

Models need machines.

◇ Estimated— Estimated: computed from our curated model and hardware catalog — not a live reading.

Modelsneedmachines.

We track the machines that matter for local AI — consumer GPUs, personal AI systems and the rentable heavy metal.

Hardware watchlist

DeviceMemorySignal
RTX 3070Last-gen 8 GB8 GBHuge install baseabout the same
RTX 4060Budget 8 GB8 GBBest-selling classabout the same
RTX 5060Current budget 8 GB8 GBCurrent genup
RTX 3060 TiLast-gen 8 GB8 GBHuge install baseabout the same
RTX 3080Last-gen 10 GB10 GBAmpere flagshipabout the same
RTX 2080 TiTuring 11 GB11 GB2018 flagshipabout the same
RTX 4070Entry local12 GBSteadyabout the same
RTX 3060 12GBBudget classic12 GBHuge install baseabout the same
RTX 5070Current 12 GB12 GBCurrent genup
RTX 4070 SuperLast-gen 12 GB12 GBAda refreshabout the same
RTX 4070 TiLast-gen 12 GB12 GBAdaabout the same
RTX 3080 TiAmpere 12 GB12 GBAmpere flagship classabout the same
RTX 4080Solid local16 GBSteadyabout the same
RTX 5080Current 16 GB16 GBCurrent genup
RTX 4060 Ti 16GBBudget 16 GB16 GBValue pickabout the same
RTX 5060 Ti 16GBCurrent budget 16 GB16 GBCurrent genup
RTX 5070 TiFast 16 GB16 GBCurrent genup
Mac mini M6 16GBApple unified 16 GB16 GB ULaunched Sep 2026up
RTX 4070 Ti SuperLast-gen 16 GB16 GBAda refreshabout the same
RX 9070 XTCurrent AMD 16 GB16 GBCurrent genup
RTX 3090Used-market value24 GBDemand risingup
RTX 4090Local workhorse24 GBDemand risingup
Mac mini M5 Pro 24GBApple unified 24 GB24 GB ULaunched Sep 2026up
RX 7900 XTXAMD 24 GB24 GBAMD flagshipabout the same
RTX 3090 TiAmpere 24 GB24 GBAmpere top cardabout the same
RTX 5090Top consumer32 GBDemand risingup
Mac Studio M5 Max 36GBApple unified 36 GB36 GB ULaunched Sep 2026up
Mac mini M4 Pro 64GBApple unified 64 GB64 GB UApple Siliconup
Mac Studio M4 Max 64GBApple unified 64 GB64 GB UApple Siliconup
Mac Studio M3 Ultra 96GBApple unified 96 GB96 GB UApple Siliconup
Mac Studio M5 Ultra 96GBApple unified 96 GB96 GB ULaunched Sep 2026up
NVIDIA DGX SparkPersonal AI system128 GB UNew categoryup
L40S 48 GB (rental)Rental mid-tier48 GB$/h trackedabout the same
A100 80 GB (rental)Rental workhorse80 GB$/h fallingdown
H100 80 GB (rental)Rental flagship80 GB$/h trackedup
H200 141 GB (rental)Rental memory flagship141 GB$/h trackedabout the same
B200 180 GB (rental)Rental frontier180 GB$/h fallingdown

Radar Card of the day

GeForce GTX 980

In 2014 this machine shipped with 4 GB. Today 4 of 49 tracked models run on it at 8K context.

Open the card →All cards →