Hy3
The cheapest provider for Hy3 is DeepInfra at $0.23 per 1M tokens (blended), charging $0.13 per 1M input and $0.53 per 1M output tokens. 6 providers serve it; the most expensive charges 1.52× as much. Context window: 256K tokens. Prices tracked since Oct 3, 2026.
Best price
$0.23
at DeepInfra
Input / output
$0.13 / $0.53
per 1M tokens
Providers
6
1.52× spread
7 days
—
best price
30 days
—
best price
Context
256K
tokens
Providers
Active offers ranked by blended price. Bars are relative to the most expensive.
| Provider | Input | Output | Blended | vs cheapest | Context | Uptime 1d | Price since |
|---|---|---|---|---|---|---|---|
DeepInfraCheapest fp4 | $0.13 | $0.53 | $0.23 | best | |||
| fp8 | $0.132 | $0.528 | $0.231 | 0.0% | |||
| bf16 | $0.14 | $0.58 | $0.25 | +9% | |||
| $0.14 | $0.58 | $0.25 | +9% | ||||
| $0.15 | $0.64 | $0.273 | +18% | ||||
| fp8 | $0.20 | $0.80 | $0.35 | +52% |
- DeepInfraCheapest$0.23$0.13 in · $0.53 outbest256K ctx99.9% uptimefp4
- $0.231$0.132 in · $0.528 out+0%256K ctx99.9% uptimefp8
- $0.25$0.14 in · $0.58 out+9%256K ctx99.8% uptimebf16
- $0.25$0.14 in · $0.58 out+9%256K ctx99.7% uptime
- $0.273$0.15 in · $0.64 out+18%256K ctx99.8% uptime
- $0.35$0.20 in · $0.80 out+52%256K ctx99.9% uptimefp8
Blended = (3 × input + output) / 4.
Price history
Hover to compare providers at any point in time. Click a legend item to hide it.
USD per 1M tokens · each step is a price change
Not enough history yet
Tracking since Oct 3, 2026. The chart fills in as new crawls arrive every 6 hours.
Long-context pricing
Higher rates once the prompt passes a size threshold.
| Offered by | Prompt size | Input | Output |
|---|---|---|---|
| Tencent | base | $0.132 | $0.528 |
| > 0 | $0.132 | $0.528 | |
| > 0 | $0.0825 | $0.33 |
Other pricing
Cache, reasoning and per-call fees. Token rates per 1M; others per unit.
| Offered by | Cache read |
|---|---|
| DeepInfra, Tencent | $0.033 |
| GMICloud, Novita | $0.035 |
| Phala | $0.04 |
| AtlasCloud | $0.05 |
Change log
Every recorded change for this model since Oct 3, 2026.
No changes yet
Tracking since Oct 3, 2026. Price changes, new providers and delistings will show up here.
About this model
Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort:...
- Released
- Jul 6, 2026
- First tracked
- Oct 3, 2026
- Canonical slug
- tencent/hy3-20260706
- Weights
- tencent/Hy3