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
writing
- Parameters
-
14.7B
- 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
9.0GB
9.0GB min
10.0GB rec
Balanced - the usual local sweet spot
Q8_0
15.0GB
15.0GB min
17.0GB 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 818.7t/s
|
fast 491.3t/s
|
| NVIDIA DGX Station 748GB |
fast 488.8t/s
|
fast 293.3t/s
|
| 8x RTX 3090 rack (192GB) |
fast 457.6t/s
|
fast 274.6t/s
|
| 4x RTX 5090 (128GB) |
fast 438.0t/s
|
fast 262.8t/s
|
| AMD Instinct MI300X (192GB) |
fast 325.3t/s
|
fast 195.2t/s
|
| 4x RTX 4090 (96GB) |
fast 246.4t/s
|
fast 147.8t/s
|
| 2x RTX 5090 (64GB) |
fast 219.0t/s
|
fast 131.4t/s
|
| 2x RTX 3090 (48GB) |
fast 114.4t/s
|
fast 68.7t/s
|
| Single RTX 5090 (32GB) |
fast 109.5t/s
|
fast 65.7t/s
|
| RTX PRO 6000 Blackwell (96GB) |
fast 109.5t/s
|
fast 65.7t/s
|
| Mac Studio M4 Ultra 192GB |
fast 72.8t/s
|
fast 43.7t/s
|
| Mac Studio M4 Ultra 512GB |
fast 72.8t/s
|
fast 43.7t/s
|
| Single RTX 4090 (24GB) |
fast 61.6t/s
|
fast 37.0t/s
|
| MacBook Pro M5 Max 128GB |
fast 40.9t/s
|
fast 24.6t/s
|
| Single GTX 1080 Ti (11GB) |
fast 29.6t/s
|
offload
|
| Dual EPYC 9004 + 768GB DDR5-4800 |
fast 28.2t/s
|
ok 16.9t/s
|
| DGX Spark 128GB unified |
ok 16.7t/s
|
ok 10.0t/s
|
| Ryzen AI Max+ 395 128GB |
ok 15.6t/s
|
ok 9.4t/s
|
| Jetson AGX Orin 64GB |
ok 12.5t/s
|
slow 7.5t/s
|
| Epyc + 512GB DDR4-3200 + 2x RTX 3090 |
ok 12.5t/s
|
slow 7.5t/s
|
| Epyc + 512GB DDR4-2400 + 2x RTX 3090 |
ok 9.4t/s
|
slow 5.6t/s
|
| NVIDIA Jetson Orin NX 16GB |
slow 6.3t/s
|
tight
|
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 about 19 hours ago via OpenRouter. Click a column to sort.
|
Provider
|
Type
|
Input $/M
|
Output $/M
|
Cache $/M
|
Tok/s
|
Latency
|
Uptime
|
Value
|
|
NextBit
|
API
|
0.10 |
0.22 |
- |
- |
- |
100.00% |
best uptime
|
|
DeepInfra
|
API
|
0.12 |
0.24 |
- |
- |
- |
100.00% |
|
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...