AI BENCHY
AD
Track all your projects in one dashboard. Get 📊stats, 🔥heatmaps and 👀recordings in one self-hosted dashboard.
uxwizz.com
#270

Hy4 preview

Tencent Release: 2026-08-29 Tested on: 2026-08-29 14:28 tencent/hy4-preview::low
770B total (49B active)MoEOpen source
(low) (none)

Summary

Hy4 preview scores 4.3 on AI BENCHY and ranks #270. It has 1.6 reliability, a 45.5% pass rate, $1.433 total cost, and 189.47s average response time.

What makes Hy4 preview unique: It uses unusually many reasoning tokens, which can help explain its slower or more expensive runs.

Model facts

Researched on 2026-08-29

Reported
Parameters
770B total (49B active)
Architecture
MoE
Availability
Open source
License
Apache-2.0

Vendor-reported backbone counts exclude the native 10B total / 0.7B active MTP layer used for speculative decoding.

Parameter source 1 Parameter source 2 Availability source

Score

4.3

Consistency

5.0

Reliability

1.6

Total Cost (Current Price)

$1.433

Total Output Tokens

559,131

Total Input Tokens

41,319

Input Price

$0.834 / 1M

Output Price

$2.501 / 1M

Tests Correct

Wrong Tests: 19

Attempt pass rate: 45.5%

Flaky tests

13

Flaky tests had mixed outcomes across runs (at least one pass and one fail).

Response Time (avg)

189.47s

Response Time (max): 597.45s

Response Time (total): 4168.32s

API error: 15 Wrong answer: 3 No answer: 1

Hamster playing table tennis

Prompt: Create a detailed SVG illustration of a hamster playing table tennis.

#270 Hy4 preview

low
Provider returned error
Cost
$0.000
Time
1.1s
Tokens
0 tok

Run history

Tested on Score Reliability Tests Correct Total Cost Compare
2026-08-29 15:57 Re-test 5.1 3.6 $1.681 Compare
2026-08-29 14:28 Initial run 4.3 1.6 $1.433 Current run

Charts

Choose the first model, then click a second model to open a side-by-side page.

Total Output Tokens

Score vs Total Output Tokens

Quick Compare

Hy4 previewlowvsNemotron 3.5 LightninghighFree AvailableHy4 previewlowvsQwen3.6 27BnoneHy4 previewlowvsKAT-Coder-Air V2.5mediumHy4 previewlowvsQwen3.8 27BnoneHy4 previewlowvsGemma 4 26B A4BnoneFree AvailableHy4 previewlowvsDeepSeek V4 Flash 0423noneHy4 previewlowvsKAT-Coder-Air V2.5highHy4 previewlowvsLaguna S 2.1mediumFree AvailableHy4 previewlowvsQwen3.5-122B-A10BnoneHy4 previewlowvsKAT-Coder-Air V2.5low

Category Breakdown

Category Score Consistency Tests Correct
Anti-AI Tricks 4.2 1.6
Coding 3.4 1.6
Combined 2.9 5.8
Data parsing and extraction 7.3 5.8
Domain specific 3.6 7.2
General Intelligence 2.8 1.6
Instructions following 9.8 10.0
Puzzle Solving 4.3 4.1
Tool Calling 3.0 10.0
Trivia 3.0 10.0

Compared models