Hy3
Tencent · Open weights · Released Jul 6, 2026 · tencent/hy3
Best $0.13 in / $0.528 out via DeepInfra · fp4
262K contextToolsReasoningStructured outputPrompt cachingOpen weights
Available from 6 providersUSD / MTOK · PRICES 23 MINUTES AGO · SPEED 3 DAYS AGO
EST. MONTHLY COSTclick a row for detail · click headers to sort
| PROVIDER | INPUT | OUTPUT | CACHE | CONTEXT | QUANT | TTFT | TPS | SMOKE | UPTIME |
|---|---|---|---|---|---|---|---|---|---|
 DeepInfra | $0.13 | $0.53 | $0.033 | 262K | fp4 | 0.73s | 178 tok/s | 2/5 | 99.8% |
 Tencent1ST PARTY | $0.132 | $0.528 | $0.033 | 262K | fp8 | ✕ | ✕ | 2/5 | 100.0% |
 GMICloud | $0.14 | $0.58 | $0.035 | 262K | bf16 | 3.3s | 112 tok/s | 2/5 | 99.7% |
 Novita | $0.14 | $0.58 | $0.035 | 262K | — | ✕ | ✕ | 2/5 | 98.6% |
 Phala | $0.15 | $0.64 | $0.04 | 262K | — | 2.1s | 87 tok/s | 2/5 | 100.0% |
 AtlasCloud | $0.20 | $0.80 | $0.05 | 262K | fp8 | 2.8s | 100 tok/s | — | 100.0% |
TTFT/TPS: median (P50) over trailing 504h · chat10k profile (~10K in / 200 out) · measured via OpenRouter routing · ✕ = probed, endpoint returned no measurable stream · SMOKE: of 5 programmatic capability probes passed (JSON schema, instruction following, tool call, long-context, benign compliance; retry-once) · screened via OpenRouter routing · v0.2 · methodology v1.1 · measured 3 days ago
SERVE THIS MODEL AND NOT LISTED? LIST YOUR INFERENCE →
Input price history · best listed price, observed
↑ 3% SINCE SEP 8, 2026
$0.126 · SEP 8, 2026$0.13 · TODAY
Pricing variants · USD / MTok in / out
STANDARD
$0.13 / $0.53
per MTok in / out
CACHE READ $0.033
Recent changesRSS ↗
Sep 23, 2026price decreased · input0.13 → 0.0825
Sep 23, 2026price decreased · output0.528 → 0.33
Sep 19, 2026price increased · input0.105 → 0.132
Sep 19, 2026price increased · output0.435 → 0.528
Sep 15, 2026price increased · input0.07 → 0.105
Sep 15, 2026price increased · output0.29 → 0.435
Sep 11, 2026price decreased · input0.132 → 0.07
Sep 11, 2026price decreased · output0.528 → 0.29
About
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:...
Specifications
Context window262,144
Max output128,000
Modalities intext
Modalities outtext
LicenseOpen weights
ReleasedJul 6, 2026
TokenizerOther
WeightsHugging Face ↗
Related
Embed badge
[](https://modelindex.ai/models/tencent/hy3)Sources
OpenRouter API23 minutes ago
AGGREGATORHugging Face23 minutes ago
COMMUNITYModelIndex benchmarks18 days ago
MEASUREDVerified 23 minutes ago · list prices, not negotiated rates.