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.
Provider transparency docs →
coding
reasoning
chat
- Parameters
-
8.2B
- Context
- 128k
- License
- apache 2.0
- Origin
-
🇨🇳
China
- Released
- Apr 2025
Run it locally
Per-quant memory needs and a static "can you run it?" reference - no rig entry required
Q4_K_M
5.2GB
5.2GB min
6.0GB rec
Balanced - the usual local sweet spot
Q8_0
8.5GB
8.5GB min
10.0GB rec
Near-lossless
The reference hardware
22 reference configs, drawn in-house. Scroll for more.
Can you run it? - reference rigs
| Rig |
Q4_K_M |
Q8_0 |
| 4x H100 80GB (320GB) |
fast 1416.8t/s
|
fast 866.9t/s
|
| NVIDIA DGX Station 748GB |
fast 845.8t/s
|
fast 517.5t/s
|
| 8x RTX 3090 rack (192GB) |
fast 791.9t/s
|
fast 484.5t/s
|
| 4x RTX 5090 (128GB) |
fast 757.9t/s
|
fast 463.7t/s
|
| AMD Instinct MI300X (192GB) |
fast 563.0t/s
|
fast 344.5t/s
|
| 4x RTX 4090 (96GB) |
fast 426.3t/s
|
fast 260.8t/s
|
| 2x RTX 5090 (64GB) |
fast 378.9t/s
|
fast 231.9t/s
|
| 2x RTX 3090 (48GB) |
fast 198.0t/s
|
fast 121.1t/s
|
| Single RTX 5090 (32GB) |
fast 189.5t/s
|
fast 115.9t/s
|
| RTX PRO 6000 Blackwell (96GB) |
fast 189.5t/s
|
fast 115.9t/s
|
| Mac Studio M4 Ultra 192GB |
fast 126.0t/s
|
fast 77.1t/s
|
| Mac Studio M4 Ultra 512GB |
fast 126.0t/s
|
fast 77.1t/s
|
| Single RTX 4090 (24GB) |
fast 106.6t/s
|
fast 65.2t/s
|
| MacBook Pro M5 Max 128GB |
fast 70.8t/s
|
fast 43.3t/s
|
| Single GTX 1080 Ti (11GB) |
fast 51.2t/s
|
fast 31.3t/s
|
| Dual EPYC 9004 + 768GB DDR5-4800 |
fast 48.7t/s
|
fast 29.8t/s
|
| DGX Spark 128GB unified |
fast 28.9t/s
|
ok 17.7t/s
|
| Ryzen AI Max+ 395 128GB |
fast 27.1t/s
|
ok 16.6t/s
|
| Jetson AGX Orin 64GB |
fast 21.7t/s
|
ok 13.3t/s
|
| Epyc + 512GB DDR4-3200 + 2x RTX 3090 |
fast 21.7t/s
|
ok 13.3t/s
|
| Epyc + 512GB DDR4-2400 + 2x RTX 3090 |
ok 16.2t/s
|
ok 9.9t/s
|
| NVIDIA Jetson Orin NX 16GB |
ok 10.8t/s
|
slow 6.6t/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.
Or run it in the cloud
Live per-provider pricing, throughput and uptime - refreshed 16 days ago via OpenRouter. Click a column to sort.
some pricing may be stale - last verified 2026-10-09
|
Provider
|
Type
|
Input $/M
|
Output $/M
|
Cache $/M
|
Tok/s
|
Latency
|
Uptime
|
Value
|
|
Alibaba
stale
|
API
|
0.12 |
0.46 |
- |
- |
- |
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.
PRICE HISTORY
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.
Loading price history...