Qwen3 30B A3B
MoE enthusiast30.5B total params, 3B active per forward pass (MoE). Min memory based on full weight loading ~15GB.
AI-generated content marks
This model embeds text watermarks in generated text and adds C2PA provenance metadata to supported files such as .png, .jpg, and .svg. Marks can be lost through editing, screenshots, or format conversion, so their absence does not prove a file is human-made.
- 30.5B
- 128k
- apache 2.0
- 🇨🇳 China
- Apr 2025
Scores
Score per dollar
600 pts per $/M input
general_score (72) divided by cheapest input price ($0.12/M). Higher is better value. See live pricing.
Related models
Save your hardware and every model page answers the real question: will it run on your machine, and how fast?
Join free - save your rig →Run it locally
Per-quant memory needs and a static "can you run it?" reference - no rig entry required
The reference hardware
22 reference configs, drawn in-house. Scroll for more.
Can you run it? - reference rigs
| Rig | Q4_K_M | Q8_0 |
|---|---|---|
| NVIDIA Jetson Orin NX 16GB | tight | no -> cloud |
| Single GTX 1080 Ti (11GB) | offload | no -> cloud |
| 4x H100 80GB (320GB) | fast 1951.5t/s | fast 1390.3t/s |
| NVIDIA DGX Station 748GB | fast 1165.1t/s | fast 830.0t/s |
| 8x RTX 3090 rack (192GB) | fast 1090.7t/s | fast 777.1t/s |
| 4x RTX 5090 (128GB) | fast 1043.9t/s | fast 743.7t/s |
| AMD Instinct MI300X (192GB) | fast 775.5t/s | fast 552.5t/s |
| 4x RTX 4090 (96GB) | fast 587.2t/s | fast 418.3t/s |
| 2x RTX 5090 (64GB) | fast 522.0t/s | fast 371.8t/s |
| 2x RTX 3090 (48GB) | fast 272.7t/s | fast 194.3t/s |
| Single RTX 5090 (32GB) | fast 261.0t/s | tight |
| RTX PRO 6000 Blackwell (96GB) | fast 261.0t/s | fast 185.9t/s |
| Mac Studio M4 Ultra 192GB | fast 173.5t/s | fast 123.6t/s |
| Mac Studio M4 Ultra 512GB | fast 173.5t/s | fast 123.6t/s |
| Single RTX 4090 (24GB) | fast 146.8t/s | offload |
| MacBook Pro M5 Max 128GB | fast 97.6t/s | fast 69.5t/s |
| Dual EPYC 9004 + 768GB DDR5-4800 | fast 67.1t/s | fast 47.8t/s |
| DGX Spark 128GB unified | fast 39.8t/s | fast 28.3t/s |
| Ryzen AI Max+ 395 128GB | fast 37.3t/s | fast 26.6t/s |
| Jetson AGX Orin 64GB | fast 29.8t/s | fast 21.3t/s |
| Epyc + 512GB DDR4-3200 + 2x RTX 3090 | fast 29.8t/s | fast 21.3t/s |
| Epyc + 512GB DDR4-2400 + 2x RTX 3090 | fast 22.4t/s | ok 15.9t/s |
Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.
Download options
Or run it in the cloud
Live per-provider pricing, throughput and uptime - refreshed about 2 hours ago via OpenRouter. Click a column to sort.
| Provider | Type | Input $/M | Output $/M | Cache $/M | Tok/s | Latency | Uptime | Value |
|---|---|---|---|---|---|---|---|---|
|
DeepInfra
|
API | 0.12 | 0.50 | - | - | - | 100.00% | best uptime |
Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.
Detailed API pricing page + JSON endpoint →
See who runs Alibaba in production →
Inference cost over time
Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.