Hy3 vs Hy4 preview: Specs & Benchmark Comparison

Hy3 and Hy4 preview both come from Tencent, Hy3 was released in July 2026, and Hy4 preview followed a month later in August 2026. Hy4 preview is the larger model at roughly 770 billion parameters, against 295 billion for Hy3.

The two models share 17 published benchmarks. Hy4 preview leads on 16 of them, Hy3 on 1. The widest gaps are on DeepSWE, where Hy4 preview scores 64.3% against 28.0%; MathArena Apex, where Hy4 preview scores 74.2% against 38.7%. Averaged across everything we track, Hy3 sits at 57.1% and Hy4 preview at 58.7%.

CharacteristicHy3Hy4 preview
CompanyTencentTencent
Release DateJuly 6, 2026August 28, 2026
Parameters295B770B
MultimodalNoNo
Average Score57.1%58.7%
Benchmarks
DeepSWE28.0%64.3%
MathArena Apex38.7%74.2%
Toolathlon48.5%74.1%
ArXivMath52.2%66.6%
Terminal-Bench 2.171.7%85.4%
NL2Repo45.6%58.9%
APEX-Agents25.6%37.1%
SuperChem54.9%66.4%
SWE-Bench Pro57.9%65.7%
SkillsBench55.3%62.9%
WideSearch76.4%83.9%
SWE-bench Multilingual75.8%82.9%
MCP Atlas79.1%83.7%
Humanity's Last Exam (no tools, text-only)47.0%43.4%
Humanity's Last Exam (with tools, text-only)53.2%55.4%
GPQA90.4%92.3%
HorizonMath7.1%8.8%

Visual Benchmark Comparison

Hy3
Hy4 preview
DeepSWE0.3 vs 0.6
0.3
0.6
MathArena Apex0.4 vs 0.7
0.4
0.7
Toolathlon0.5 vs 0.7
0.5
0.7
ArXivMath0.5 vs 0.7
0.5
0.7
Terminal-Bench 2.10.7 vs 0.9
0.7
0.9
NL2Repo0.5 vs 0.6
0.5
0.6
APEX-Agents0.3 vs 0.4
0.3
0.4
SuperChem0.5 vs 0.7
0.5
0.7
SWE-Bench Pro0.6 vs 0.7
0.6
0.7
SkillsBench0.6 vs 0.6
0.6
0.6
WideSearch0.8 vs 0.8
0.8
0.8
SWE-bench Multilingual0.8 vs 0.8
0.8
0.8
MCP Atlas0.8 vs 0.8
0.8
0.8
Humanity's Last Exam (no tools, text-only)0.5 vs 0.4
0.5
0.4
Humanity's Last Exam (with tools, text-only)0.5 vs 0.6
0.5
0.6

Verdict

Hy4 preview leads in 1 out of 2 comparison categories.

Overall Performance

Both models show comparable average scores: Hy3 — 0.6, Hy4 preview — 0.6.

Recency

Hy4 preview is newer: released 8/28/2026 vs 7/6/2026.

More About These Models

Related Comparisons

Frequently Asked Questions

Which is better for coding — Hy3 or Hy4 preview?
Direct comparison on the SWE-Bench benchmark is not available. We recommend reviewing other metrics on the comparison page.
Which model is cheaper — Hy3 or Hy4 preview?
API pricing data is available on the individual model pages.
Which has a larger context window — Hy3 or Hy4 preview?
Context window data is available on the individual model pages.

The Hy3 and Hy4 preview comparison is updated for 2026. Data includes benchmark results, API pricing, context window size and other specifications. For more detailed information, visit the Hy3 or Hy4 preview page. See also the complete list of AI model comparisons.