Qwen-Image-2.1-mflux-q4 ↗
apache-2.0
Repository updated: 2026-09-26FIND YOUR FIT
Find the right local AI model for your task
Choose what you want to do. Explore quality, hardware estimates and sources before deciding what to run.
QUALITY × LOCAL MEMORY
Which open models lead?
| # | Model | LiveBench / 100 | Memory · Q4 estimate | Weight license |
|---|---|---|---|---|
| 1 | Smaug MiniAbacus AI | 76.9 | ~29 GB | Apache 2.0 ↗ |
| 2 | Qwen3.8 27BAlibaba | 75.3 | ~29 GB | Apache 2.0 ↗ |
| 3 | Qwen3.6 27BAlibaba | 64.0 | ~29 GB | Apache 2.0 ↗ |
LiveBench 2026-06-25 ↗ · Selected task: Overall · Scores collected: 2026-09-26. Positions cover only evaluated models within the selected memory budget. This LiveBench score is not the Artificial Analysis Intelligence Index, and the two scales are not directly comparable.
Speed depends on the computer, runtime and quantization. Check tokens/s and time to first token with the measured configuration.
Checking the latest saved data…
LOCAL AI WIZARD
What do you want to do?
Start with the task. Find the right model for you.
FROM THE CATALOG
Recently updated repositories
These entries update with the catalog, even without a benchmark. Repository changes are not necessarily new model releases.
Qwen3.6-35B-A3B-DFlash2-4bit ↗
apache-2.0
Repository updated: 2026-09-25Qwen3.8-Flash-Next-4bit-mtp ↗
Check license at source
Repository updated: 2026-09-25GLM-5.3-Flash-4bit ↗
mit
Repository updated: 2026-09-25NV-Reason-CT ↗
openmdw-1.1
Repository updated: 2026-09-25Ling-mini-2.0 ↗
mit
Repository updated: 2026-09-24Local image models
Image models
51 open-weight entries in the Artificial Analysis snapshot checked 2026-09-22, with deeper technical review where primary evidence is available.
i
How to read this catalog
The complete ranked set comes from one Artificial Analysis arena snapshot per modality, so ranks and Elo scores are comparable only inside that modality. API variants can share the same public checkpoint.
A reviewed badge means license, components or hardware were checked against primary sources. Benchmark-only entries remain visible for coverage, with unknown facts clearly marked instead of inferred.
OPEN-WEIGHT LEADERS
Top image models in this snapshot
Search, filters & my hardware 51/51
51 models · 51 ranked · 9 technical profiles reviewed · benchmark snapshot ↗ 2026-09-22
QUALITY × HARDWARE
Benchmark score vs documented GPU memory
Standard pathQuantization or offload
Top ranked · memory unverifiedTheir scores are known, but there is no documented GPU-memory value for the horizontal axis.
19 of 51 visible models have a documented memory path. Open the rankings for every model.
SIDE BY SIDE
0/3Compare selected models
| Field |
|---|
| Task |
| Parameters |
| Repository size |
| Components |
| Hardware |
| License |
Choose models to compare.
Local video models
Video models
7 open-weight entries in the Artificial Analysis snapshot checked 2026-09-22, with deeper technical review where primary evidence is available.
i
How to read this catalog
The complete ranked set comes from one Artificial Analysis arena snapshot per modality, so ranks and Elo scores are comparable only inside that modality. API variants can share the same public checkpoint.
A reviewed badge means license, components or hardware were checked against primary sources. Benchmark-only entries remain visible for coverage, with unknown facts clearly marked instead of inferred.
OPEN-WEIGHT LEADERS
Top video models in this snapshot
Search, filters & my hardware 13/13
13 models · 7 ranked · 6 technical profiles reviewed · benchmark snapshot ↗ 2026-09-22
QUALITY × HARDWARE
Benchmark score vs documented GPU memory
Standard pathQuantization or offload
Top ranked · memory unverifiedTheir scores are known, but there is no documented GPU-memory value for the horizontal axis.
2 of 13 visible models have a documented memory path. Open the rankings for every model.
SIDE BY SIDE
0/3Compare selected models
| Field |
|---|
| Task |
| Parameters |
| Repository size |
| Components |
| Hardware |
| License |
Choose models to compare.
Local audio models
Audio models
16 open-weight entries in the Artificial Analysis snapshot checked 2026-09-22, with deeper technical review where primary evidence is available.
i
How to read this catalog
The complete ranked set comes from one Artificial Analysis arena snapshot per modality, so ranks and Elo scores are comparable only inside that modality. API variants can share the same public checkpoint.
A reviewed badge means license, components or hardware were checked against primary sources. Benchmark-only entries remain visible for coverage, with unknown facts clearly marked instead of inferred.
OPEN-WEIGHT LEADERS
Top audio models in this snapshot
Search, filters & my hardware 19/19
19 models · 16 ranked · 3 technical profiles reviewed · benchmark snapshot ↗ 2026-09-22
QUALITY × HARDWARE
Benchmark score vs documented GPU memory
Standard pathQuantization or offload
Top ranked · memory unverifiedTheir scores are known, but there is no documented GPU-memory value for the horizontal axis.
8 of 19 visible models have a documented memory path. Open the rankings for every model.
SIDE BY SIDE
0/3Compare selected models
| Field |
|---|
| Task |
| Parameters |
| Repository size |
| Components |
| Hardware |
| License |
Choose models to compare.
Find models for your hardware
Choose a device and memory capacity. Each profile brings together specifications, candidate models, settings and available measurements.
Mac mini
There was no Mac mini M3. Compare generations, then choose the real memory configuration.
Mac mini M6 · 2026
Choose chip and memory. Each option opens its profile and model estimates for that capacity.
Mac mini M5 Pro · 2026
Choose chip and memory. Each option opens its profile and model estimates for that capacity.
Mac mini M4 Pro · 2024
Choose chip and memory. Each option opens its profile and model estimates for that capacity.
Mac mini M4 · 2024
Choose chip and memory. Each option opens its profile and model estimates for that capacity.
NVIDIA RTX
Card VRAM; system RAM and runtime support are separate. Check the SKU when 8 and 16 GB versions exist.
Other computers
Mac Studio
Choose chip and memory. Each option opens its profile and model estimates for that capacity.
Your hardware is not listed?
SOURCE-BACKED GUIDES
Take a closer look
12 checkpoint profiles, 22 hardware configurations and 3 task comparisons. Each page keeps its source, date and limits visible.
Reviewed model profiles
All discovered models →Qwen3.5 9Btext + image + videoQwen3.8 27Btext + image + videoQwen3.5 35B A3Btext + image + videoQwen3.6 27Btext + imageSmaug Minitext + imageDeepSeek V4 Flash · 0731textGLM-5.3 Flashtext + imageKimi K2.6 · thinkingtext + imageFLUX.2 [dev]imageQwen-ImageimageWan2.1 T2V 1.3BvideoStable Audio Open SmallaudioHardware
Mac mini M4 · 16 GB unifiedMac mini M4 Pro · 48 GB unifiedMac mini M6 · 16 GB unifiedMac mini M6 · 24 GB unifiedMac mini M6 · 32 GB unifiedMac mini M5 Pro · 48 GB unifiedRTX 4060 · 8 GB VRAMRTX 5060 Ti · 16 GB VRAMRTX 5080 · 16 GB VRAMFramework Desktop · Ryzen AI Max+ 395 · 128 GBMac Studio M5 Max · 128 GB unifiedRTX 3060 · 12 GB VRAMRTX 4060 Ti · 16 GB VRAMRTX 4070 Ti SUPER · 16 GB VRAMRTX 4090 · 24 GB VRAMMac mini M4 · 24 GB unifiedMac mini M4 · 32 GB unifiedRTX 5090 · 32 GB VRAMMac mini M5 Pro · 64 GB unifiedDGX Spark · 128 GB unifiedMac Studio M5 Ultra · 256 GB unifiedDGX H100 · 8 × 80 GBData and methodologySources, methodology & data status
Benchmark release: 2026-06-25Data checked: 2026-09-26Verified snapshot · checking source
1420 source repositories in the catalog.
01 / LiveBench
The overall score is the equally weighted mean of seven category means (0–100). All scores use one release. This is not the Artificial Analysis Intelligence Index; the two scales must not be mixed.
CSV ↗ · Categories ↗ · LiveBench ↗ · Data reuse policy ↗02 / Memory estimate
Planning formula: total parameters × 0.625 bytes, plus 20% working allowance and 8 GB reserve, rounded up in decimal GB. This is a heuristic, not a measured minimum or guarantee. Quantization, context, KV cache, vision components, runtime and offloading change requirements. MoE uses total parameters, not only active ones.
Qwen counts refer to language parameters; GLM-5.3 and DeepSeek V4 use repository tensor totals; GLM-5.3 Flash uses the manufacturer’s 320B specification. Architectures without a reviewed estimate remain outside the memory axis.
03 / Reference, not a local test
LiveBench evaluates the named model and reasoning effort. Its score is not a measurement of the local Q4 variant. Hardware links are official destinations without affiliate tracking. Stock, exact configuration and regional availability can change.
04 / Automatic updates without paid APIs
The deployed product checks scores, approved official model organizations, US prices and stock daily, and hardware specifications weekly. Due visits can also refresh safely. D1 preserves the last valid snapshot. Incompatible methodology is quarantined; unavailable sources never erase valid data. New models remain pending benchmark or insufficient data until official facts exist.