Compare / head-to-head
Hy4 previewvs
Step 3.7 Flash
Hy4 preview leads 4 of 4 shared benchmarks. Hy4 preview has the larger context window (1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
4 – 0
Hy4 preview leads
Cheaper per token
—
list price, input + output
Larger context
Hy4 preview
1M tokens
Providers
2 providers
Tencent Hunyuan · StepFun
Hy4 preview
Tencent Hunyuan · released 2026-08-28
text
Step 3.7 Flash
StepFun · released 2026-05-29
textvision
- Context
- 256K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 19 · 6 core
Quality
Benchmark matrix
4 shared · 14 only Hy4 preview · 15 only Step 3.7 Flash| Benchmark | Hy4 preview | Step 3.7 Flash | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 4 | ||||
| HLE (with tools, text-only) | 55.4% ↗ | 49.7% ↗ | +5.7 pt | Hy4 preview |
| SWE-bench Multilingual | 82.9% ↗ | 72.4% ↗ | +10.5 pt | Hy4 preview |
| SWE-bench Pro | 65.7% ↗ | 56.3% ↗ | +9.4 pt | Hy4 preview |
| Terminal-Bench 2.1 | 85.4% ↗ | 59.6% ↗ | +25.8 pt | Hy4 preview |
| Only Hy4 preview reports · 14 | ||||
| GPQA Diamond | 92.3% ↗ | not reported | — | — |
| Agents' Last Exam (ALE-CLI) | 22.8% ↗ | not reported | — | — |
| CyberGym | 78.4% ↗ | not reported | — | — |
| DeepSWE | 64.3% ↗ | not reported | — | — |
| HLE (no tools, text-only) | 43.4% ↗ | not reported | — | — |
| LMArena Agent (LMArena) | 0.0395 ↗ | not reported | — | — |
| LMArena WebDev (LMArena) | 1633.2 ↗ | not reported | — | — |
| MathArena Apex 2025 | 74.2% ↗ | not reported | — | — |
| MCP-Atlas (public) | 83.7% ↗ | not reported | — | — |
| NL2Repo-Bench | 58.9% ↗ | not reported | — | — |
| OfficeQA Pro | 66.2% ↗ | not reported | — | — |
| Terminal-Bench 4.0 (Vals AI) | 8.1% ↗ | not reported | — | — |
| Toolathlon-Verified | 74.1% ↗ | not reported | — | — |
| WideSearch | 83.9% ↗ | not reported | — | — |
| Only Step 3.7 Flash reports · 15 | ||||
| SWE-bench Verified | not reported | 76.5% ↗ | — | — |
| AIME 2026 | not reported | 95% ↗matharena ⚠ | — | — |
| AA-LCR | not reported | 63.9% ↗ | — | — |
| BrowseComp | not reported | 75.8% ↗ | — | — |
| ClawEval v1.1 | not reported | 67.1% ↗ | — | — |
| DeepSearchQA Accuracy | not reported | 81.7% ↗ | — | — |
| DeepSearchQA F1 | not reported | 92.8% ↗ | — | — |
| GDPval-Stirrup | not reported | 1415.8 ↗ | — | — |
| GDPval-Stirrup Intelligence Index | not reported | 45.8 ↗ | — | — |
| HLE (with tools) | not reported | 47.2% ↗ | — | — |
| ResearchRubrics | not reported | 71.7% ↗ | — | — |
| SimpleVQA (with tools) | not reported | 79.2% ↗ | — | — |
| Toolathlon | not reported | 49.5% ↗ | — | — |
| V* (with Python) | not reported | 95.3% ↗ | — | — |
| WorldVQA (with visual search) | not reported | 58.1% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Hy4 preview minus Step 3.7 Flash in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Hy4 preview | Step 3.7 Flash | Edge |
|---|---|---|---|
| Context window | 1M tokens | 256K tokens | Hy4 preview |
| Max output | 64K tokens | — | — |
| Input price / 1M | $0.834 ↗ | — | — |
| Output price / 1M | $2.5 ↗ | — | — |
| Cached input / 1M | $0.042 ↗ | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $3.34 | — | — |
| Modalities | text | text · vision | Step 3.7 Flash |
| Released | 2026-08-28 | 2026-05-29 | — |
| Cited benchmark scores | 18 | 19 | — |
Reliability
Provider status
Live from /statusMore matchups
Hy4 preview vs …
Models sharing the most benchmarksMore matchups
Step 3.7 Flash vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Hy4 preview and Step 3.7 Flash on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.