AI Compute RadarWhat runs on my machine?
Pick your card. 49 models, one word each.
Every tracked model gets one of six words for your card and the memory it needs. Runs in your browser; nothing is sent.
Files measured · context computed · no benchmarks
Field guide — Issue 063Monday · 2 Nov 2026 · Models + compute decision layer
Three questions.Three 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 does a GPU hour cost today?
RTX 4090 from $0.34
$0.50 typical on the Vast marketplace
Down from last week. Eight classes, ten sources, every price read from the public page. Nothing is blended.
Collected 33 min ago · RunPod list price · Vast middle of 64 offers
What is popular this week?
Pick of the week: Gemma 4 E4B.
8 billion on disk, under a gigabyte of context at 32K
- Qwen3 32Bmore popular ▲
- Gemma 3 27Bmore popular ▲
Popularity on Hugging Face · Collected 1 h ago
- ◆LFM3 1.7B · HF trending 686
- ▲Qwen3.8 27B · HF trending 283
- ▲Qwen3.8 Flash-Next · HF trending 115
- ▼Llama 3.1 8B · HF trending 58
- ▲RTX 3090 $0.27/h
- ▲RTX 4090 $0.50/h
- ◆A100 $0.73/h
- ◆RTX 5090 $0.73/h
- ◆L40S $0.80/h
- ◆H100 $3.24/h
- ◆H200 $5.88/h
- ◆B200 $8.75/h
Today’s edition — Issue 063 · Monday · 2 Nov 2026
Since the previous edition · newest reading 2026-11-02 16:33 UTC
- More popular+8Qwen3 32B · 34 → 42 · popularity2026-11-01 18:15 UTC → 2026-11-02 16:33 UTC
- New on the radar this week3LFM3 1.7B, Ornith 1.5 122B, LFM2.5 24B-A2B2026-10-28 18:15 UTC → 2026-11-02 15:30 UTC
- GPU rent down−12.8%A100 · $0.99 → $0.87 per hour · Vast.ai daily median2026-10-31 23:00 UTC → 2026-11-01 23:00 UTC
- Trending up+7Gemma 4 E4B · 28 → 35 · Hugging Face trending2026-11-01 18:15 UTC → 2026-11-02 15:30 UTC
- More popular+7Gemma 3 27B · 50 → 57 · popularity2026-11-01 18:15 UTC → 2026-11-02 16:33 UTC
Pick of the week — Week 44, 2026
Gemma 4 E4B
Gemma 4 E4B by Google takes the eighth pick: Heat Score 59 to 76 between the readings of 24 and 30 October, the largest gain among models that run comfortably on a 24 GB card and are not on cooldown. Its Q4_K_M file measures 4.6 GB. At 8K context it needs 5.8 GB, it runs on the 8 GB cards and runs easily from 10 GB up; at 32K it needs 7.2 GB, and only 0.9 GB of that is context cache, a fifth of what last week's pick needs at the same length.
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.
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
- Ornith 1.5 122BOrnith AI · repo created 2026-10-16 · 17 d agocollecting
- LFM2.5 24B-A2BLiquid AI · repo created 2026-10-12 · 21 d agocollecting
- GLM-5.3Zhipu AI · repo created 2026-08-25 · 69 d agoHeat 73
- GLM-5.3-FlashZhipu AI · repo created 2026-08-25 · 69 d agoHeat 73
- Qwen3.8 Flash-NextAlibaba Qwen · repo created 2026-08-24 · 70 d agoHeat 78
Strongest returns
- Gemma 4 E4BGoogle · 245 d ago82
- Qwen3.8 Flash-NextAlibaba Qwen · 70 d ago78
- Qwen3.8 27BAlibaba Qwen · 89 d ago76
Not plotted, no dated open-weights repository: LFM3 1.7B.
Ages come from Hugging Face repository creation dates; popularity from measured signals. How popularity is scored
01 / Model market
What is winning
| Rank | Model | Heat0–100 · composite | TrendingHugging Face | DownloadsHugging Face | $/M inOpenRouter |
|---|---|---|---|---|---|
| 1 | Gemma 4 E4BGoogle · OpenGeneralHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 35 | 35 | 4.1M+3.8% up | — | |
| 2 | Qwen3.8 Flash-NextAlibaba Qwen · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 115$/M in: $0.15 | 115 | 3.9M+24.3% up | $0.15 | |
| 3 | Qwen3.8 27BAlibaba Qwen · OpenGeneral · Coding · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 283$/M in: $0.15 | 283 | 7.9M+4.0% up | $0.15 | |
| 4 | gpt-oss 20BOpenAI · OpenGeneral · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 20$/M in: $0.03 | 20 | 5.9M+2.7% up | $0.03 | |
| 5 | Llama 3.1 8BMeta · OpenGeneral · CodingHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 58 | 58 | 4.5M-7.9% down | — | |
| 6 | GLM-5.3-FlashZhipu AI · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 54$/M in: $0.03 | 54 | 6.3M-0.6% down | $0.03 | |
| 7 | GLM-5.3Zhipu AI · OpenGeneral · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 40$/M in: $0.08 | 40 | 2.4M+14.6% up | $0.08 | |
| 8 | Gemma 4 31BGoogle · OpenGeneral · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 41$/M in: $0.20 | 41 | 8.1M-2.6% down | $0.20 | |
| 9 | DeepSeek-R1DeepSeek · OpenResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 12 | 12 | 1.7M+12.9% up | — | |
| 10 | gpt-oss 120BOpenAI · OpenGeneral · Research · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 27$/M in: $0.15 | 27 | 2.5M-13.3% down | $0.15 | |
| 11 | Ornith 1.5 35B-A3BOrnith AI · OpenGeneral · Coding · AgentsHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 18 | 18 | 276K+1.6% up | — | |
| 12 | Qwen3.8 2.4T-A95BAlibaba Qwen · OpenGeneral · Agents · ResearchHugging Face ↗ (External link)GGUF ↗ (External link)Trending: 17$/M in: $2.00 | 17 | 26K+3.5% up | $2.00 |
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$0.27up
- RTX 4090$0.50up
- A100$0.73about the same
- RTX 5090$0.73about the same
- L40S$0.80about the same
- H100$3.24about the same
- H200$5.88about the same
- B200$8.75about the same
03 / My setup
Your machine, your answer
◇ Estimated— Estimated: computed from our curated model and hardware catalog — not a live reading.
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 44, 2026
Gemma 4 E4B is the pick; LFM2.5 8B-A1B 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
| Device | Memory | Class | Signal |
|---|---|---|---|
| RTX 3070Last-gen 8 GB | 8 GB | Huge install baseabout the same | |
| RTX 4060Budget 8 GB | 8 GB | Best-selling classabout the same | |
| RTX 5060Current budget 8 GB | 8 GB | Current genup | |
| RTX 3060 TiLast-gen 8 GB | 8 GB | Huge install baseabout the same | |
| RTX 3080Last-gen 10 GB | 10 GB | Ampere flagshipabout the same | |
| RTX 2080 TiTuring 11 GB | 11 GB | 2018 flagshipabout the same | |
| RTX 4070Entry local | 12 GB | Steadyabout the same | |
| RTX 3060 12GBBudget classic | 12 GB | Huge install baseabout the same | |
| RTX 5070Current 12 GB | 12 GB | Current genup | |
| RTX 4070 SuperLast-gen 12 GB | 12 GB | Ada refreshabout the same | |
| RTX 4070 TiLast-gen 12 GB | 12 GB | Adaabout the same | |
| RTX 3080 TiAmpere 12 GB | 12 GB | Ampere flagship classabout the same | |
| RTX 4080Solid local | 16 GB | Steadyabout the same | |
| RTX 5080Current 16 GB | 16 GB | Current genup | |
| RTX 4060 Ti 16GBBudget 16 GB | 16 GB | Value pickabout the same | |
| RTX 5060 Ti 16GBCurrent budget 16 GB | 16 GB | Current genup | |
| RTX 5070 TiFast 16 GB | 16 GB | Current genup | |
| Mac mini M6 16GBApple unified 16 GB | 16 GB U | Launched Sep 2026up | |
| RTX 4070 Ti SuperLast-gen 16 GB | 16 GB | Ada refreshabout the same | |
| RX 9070 XTCurrent AMD 16 GB | 16 GB | Current genup | |
| RTX 3090Used-market value | 24 GB | Demand risingup | |
| RTX 4090Local workhorse | 24 GB | Demand risingup | |
| Mac mini M5 Pro 24GBApple unified 24 GB | 24 GB U | Launched Sep 2026up | |
| RX 7900 XTXAMD 24 GB | 24 GB | AMD flagshipabout the same | |
| RTX 3090 TiAmpere 24 GB | 24 GB | Ampere top cardabout the same | |
| RTX 5090Top consumer | 32 GB | Demand risingup | |
| Mac Studio M5 Max 36GBApple unified 36 GB | 36 GB U | Launched Sep 2026up | |
| Mac mini M4 Pro 64GBApple unified 64 GB | 64 GB U | Apple Siliconup | |
| Mac Studio M4 Max 64GBApple unified 64 GB | 64 GB U | Apple Siliconup | |
| Mac Studio M3 Ultra 96GBApple unified 96 GB | 96 GB U | Apple Siliconup | |
| Mac Studio M5 Ultra 96GBApple unified 96 GB | 96 GB U | Launched Sep 2026up | |
| NVIDIA DGX SparkPersonal AI system | 128 GB U | New categoryup | |
| L40S 48 GB (rental)Rental mid-tier | 48 GB | $/h trackedabout the same | |
| A100 80 GB (rental)Rental workhorse | 80 GB | $/h fallingdown | |
| H100 80 GB (rental)Rental flagship | 80 GB | $/h trackedup | |
| H200 141 GB (rental)Rental memory flagship | 141 GB | $/h trackedabout the same | |
| B200 180 GB (rental)Rental frontier | 180 GB | $/h fallingdown |
Radar Card of the day
Mac mini M4 Pro 64GB
In 2024 this machine shipped with 64 GB. Today 31 of 49 tracked models run on it at 8K context.