LLM Comparison

Hy4 Preview vs Hy3: benchmark scores, pricing & comparison.

Side-by-side Hy4 Preview vs Hy3 comparison across SWE-bench, GPQA, HLE, Terminal-Bench, coding agent scores, token pricing, context window, and AskClash RWT. Green marks the winner on each benchmark.

Open live leaderboardHy4 Preview model pageHy3 model page
Rank #18 vs #32AskClash overall scores 58.2 vs 35.1.
Pricing $0.83/$2.50 vs $0.12/$0.43Input and output token prices per 1M tokens when published.
Open Weight vs Open WeightTencent vs Tencent.

Hy4 Preview vs Hy3 benchmark comparison

Green cells highlight the winning model for each metric. Scores are cached from the AskClash LLM leaderboard snapshot.

MetricHy4 PreviewHy3
Overall Score58.235.1
Leaderboard Rank#18#32
RWT8.58.0
HLE43.453.2
GPQA92.390.4
SWE-bench—78.0
SWE-Pro65.757.9
Terminal-Bench85.471.7
DeepSWE64.3—
GDPval-AA1678.0—
MCP Atlas83.779.1
Tau2—92.7
MRCR—75.1
Input Price (per 1M tokens)$0.83$0.12
Output Price (per 1M tokens)$2.50$0.43
Context Window1M256K
Benchmarks Published79

Hy4 Preview vs Hy3 head-to-head charts

Hy4 Preview leads 6 and Hy3 leads 1 of 7 shared benchmarks. Hy3 is cheaper on a defined 1M-input / 200K-output workload. Charts show only benchmarks both models publish.

Hy4 PreviewHy3
Overall
58.2Hy4 Preview
35.1Hy3
RWT
8.5Hy4 Preview
8.0Hy3
HLE
43.4Hy4 Preview
53.2Hy3
GPQA
92.3Hy4 Preview
90.4Hy3
SWE-Pro
65.7Hy4 Preview
57.9Hy3
Terminal-Bench
85.4Hy4 Preview
71.7Hy3
MCP Atlas
83.7Hy4 Preview
79.1Hy3
Hy4 Preview
Input$0.83
Output$2.50
Workload$1.33
Context1M
Hy3
Input$0.12
Output$0.43
Workload$0.21
Context256K

Workload = published cost of 1M input + 200K output tokens. Open the live leaderboard for interactive compare charts.

More Hy4 Preview and Hy3 comparisons

Explore how Hy4 Preview and Hy3 stack up against other top-ranked LLMs.

Claude Opus 5 vs Hy4 PreviewClaude Opus 5 vs Hy3Claude Fable 5 vs Hy4 PreviewClaude Fable 5 vs Hy3GLM-5.3 vs Hy4 PreviewGLM-5.3 vs Hy3Grok 4.6 vs Hy4 PreviewGrok 4.6 vs Hy3GPT-5.6 Sol vs Hy4 PreviewGPT-5.6 Sol vs Hy3

How to read this comparison

Benchmark scores

Higher is better for all benchmark scores (SWE-bench, GPQA, HLE, Terminal-Bench, etc.). Green marks the model with the higher score.

Token pricing

Lower is better for input and output prices. Green marks the cheaper model per 1M tokens.

Coverage matters

Models with fewer disclosed benchmark cells may have inflated percentile scores. Check the benchmark cell count for context.

This comparison page is generated from the AskClash LLM leaderboard cache. Open the live leaderboard for real-time scores and interactive filtering.