OpenAI compatible API. Attested gateway. Public status.

Tencent: Hy3 preview Benchmarks

Benchmark and measurement links for Tencent: Hy3 preview, with TrustedRouter route data first.

Verify gateway
1 URLbase_url migration
100smodels and routes
0prompt logs by default

tencent/hy3-preview

Benchmarks

All models

AI IQ IQ 100 #50 public AI IQ rank for hy3
View AI IQ profile

Published benchmark scores

Benchmark scores for Tencent: Hy3 preview — every row links to its source, and a score is only ever attached to the exact checkpoint it was measured on. Vendor model-card and open-leaderboard numbers are cited, not run by us. Rows marked TrustedRouter · replays published are our own runs of this model through the gateway, with the full per-item replay published in trustedrouter-benchmarks so anyone can re-grade them.

BenchmarkCategoryScoreSource
Aider Polyglot
34 Exercism exercises (Python), pass@1, real unit tests (no judge)
 Coding 41.2% TrustedRouter Benchmarks replay
2026-06-18
SimpleQA Verified
250 closed-book questions, no tools; GPT-4.1 autorater (Google's exact prompt); 32768-token budget
 Factuality 28.0% TrustedRouter Benchmarks replay
2026-06-18
IFEval
100-prompt subset, 0-shot; Google's deterministic verifiers (no judge); score = avg of strict/loose x prompt/instruction
 Instruction following 91.2% TrustedRouter Benchmarks replay
2026-06-18
MMLU-Pro
200-question stride-sampled subset (TIGER-Lab/MMLU-Pro), 10-choice CoT, letter-match; no judge
 Knowledge 76.0% TrustedRouter Benchmarks replay
2026-06-18
GSM8K
30-problem subset, deterministic numeric match (no judge); near-saturated, kept as a sanity check
 Math 100.0% TrustedRouter Benchmarks replay
2026-06-18

TrustedRouter measurements

TrustedRouter publishes route and status measurements without storing prompt or output content. Provider latency and uptime are exposed through the model performance and uptime pages.

External benchmark references

Sign in

Choose a sign in method.