what/llm/can/i/run
Scanning your GPU...
Hardware-first local LLM ranker

Pick the right local LLM for the GPU you already own.

Tell us what's in your machine, and we'll rank every model that fits — by real benchmarks (LiveBench, Aider, Arena), with the math shown. No fluff, no upsell.

Explore by GPU tier

4 GB iGPU
Intel Iris XeIntel Iris Xe · OEM laptop
Top pickGemma 3 4BQ6_K · 3 fit
12 GB
RTX 3060 12GBRTX 3060 · RTX 4070
Top pickgemma 3 27bQ2_K · 7 fit
16 GB
RTX 4080 SUPERRTX 4080 Super · RTX 4060 Ti 16GB
Top pickgemma 3 27bQ4_K_M · 12 fit
24 GB
RTX 4090RTX 4090 · RTX 3090
Top pickgemma 3 27bQ6_K · 13 fit
32 GB
RTX 5090RTX 5090
Top pickgemma 3 27bQ8_0 · 16 fit
48 GB workstation
A6000 ProA6000 Pro
Top pickgemma 3 27bQ8_0 · 17 fit
64 GB unified
M3 Max 64GBM3 Max 64GB · M2 Max
Top pickQwen2.5 Coder 32BQ8_0 · 18 fit
128 GB unified
M3 Max 128GBM3 Max 128GB · Mac Studio
Top pickgemma 3 27bFP16 · 19 fit
Want personalized ranking?
Tell us your exact GPU — we'll rank every model that fits with quant, tok/s, and the math shown.
Pick my GPU →
Also poke around
CalculatorVRAM math
Plug in any model + quant + context window. Get exact VRAM (weights + KV cache + activation + overhead), tok/s estimate, and which GPUs fit. KV cache often dominates at 32k+ — most calculators get this wrong.Use when: Planning a new build · deciding context length
CompareGPU vs GPU
Side-by-side cards with spec diff, unlocked models, tok/s deltas. The honest answer to "should I upgrade from 4090 to 5090?" — usually no.Use when: Buying decision · upgrade timing

▸ Data provenance & freshness

Site data bundled 2026-05-19
SourceFieldsSource dateStatus
Aider Polyglotpolyglot pass_rate · total_cost_usd · edit_format2025-11✓ Fresh
LiveBench
Scores aggregated client-side from per-question parquet; verify against livebench.ai if numbers differ by >2%.
Reasoning · Coding · Math · Data · Language · IFEval (aggregated to overall)2026-01-08✓ Fresh
Chatbot Arena (legacy MT-bench/MMLU)
The HF Space CSVs are the old leaderboard format — NOT current Arena ELO. Real ELO scraper (fboulnois mirror) lands in next refresh.
MT-bench · MMLU · License · Organization · Link2025-08-04⚠ Stale
HuggingFace APImodel_id · downloads · params · context · license · tagslive✓ Live
Open LLM Leaderboard v2
Project effectively abandoned · we dropped this source.
(not used — project archived)2024-08-07✗ Dropped

Why some sources lag: real-time benchmark scraping is on the v1.1 roadmap (GitHub Actions daily cron). Until then we ship a known-good snapshot and mark its date so you can compare against fresher sources yourself.