Model comparison
GPT-5.4 mini vs Step 3.7 Flash
Head-to-head evidence from 20 shared benchmark results across 6 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
BenchAlign evidence: GPT-5.4 mini estimated; Step 3.7 Flash estimated. Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. GPT-5.4 mini and Step 3.7 Flash share 20 comparable benchmark results. 1 of 8 categories are comparable. 11 results are unique to GPT-5.4 mini; 10 to Step 3.7 Flash.
Updated July 16, 2026- Shared results
- 20
- GPT-5.4 mini only
- 11
- Step 3.7 Flash only
- 10
- Comparable categories
- 1 / 8
Pick GPT-5.4 mini if you want the stronger benchmark profile. Step 3.7 Flash only becomes the better choice if agentic is the priority or you want the cheaper token bill.
Confidence note. This is a partial-evidence comparison with 20 shared benchmark results across 6 evidence categories; 1 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
GPT-5.4 mini is clearly ahead on the provisional aggregate, 64 to 57. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GPT-5.4 mini is also the more expensive model on tokens at $0.75 input / $4.50 output per 1M tokens, versus $0.20 input / $1.15 output per 1M tokens for Step 3.7 Flash. That is roughly 3.9x on output cost alone. GPT-5.4 mini gives you the larger context window at 400K, compared with 256K for Step 3.7 Flash.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | GPT-5.4 mini | Δ | Step 3.7 Flash |
|---|---|---|---|
| Agentic | GPT-5.4 mini65.7 | Margin→ 0.7 | Step 3.7 Flash66.4 |
| Coding | GPT-5.4 miniNot measured | MarginNo overlap | Step 3.7 Flash56.3 |
| Knowledge | GPT-5.4 mini47.8 | MarginNo overlap | Step 3.7 FlashNot measured |
| Math | GPT-5.4 mini21.7 | MarginNo overlap | Step 3.7 FlashNot measured |
| Multimodal | GPT-5.4 mini76.6 | MarginNo overlap | Step 3.7 FlashNot measured |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
Terminal-Bench 2.0
AgenticA 60%B 59.5%Winner: GPT-5.4 miniΔ 0.5Terminal-Bench 2.0: GPT-5.4 mini scored 60%; Step 3.7 Flash scored 59.5%. GPT-5.4 mini wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | GPT-5.4 mini | Step 3.7 Flash | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | GPT-5.4 mini$0.75 input / $4.5 output | Step 3.7 Flash$0.2 input / $1.15 output | Step 3.7 Flash has the lower combined listed price. |
| Generation speedtokens per second | GPT-5.4 mini201 tok/s | Step 3.7 FlashNot available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | GPT-5.4 mini3.85 s | Step 3.7 FlashNot available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | GPT-5.4 mini400K | Step 3.7 Flash256K | GPT-5.4 mini lists the larger context window. |
Benchmark Deep Dive
AgenticStep 3.7 Flash wins14 benchmarks
| Benchmark | GPT-5.4 mini | Step 3.7 Flash | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 60% | 59.5% | GPT-5.4 mini leads |
| OSWorld-VerifiedSource | 72.1% | — | Not comparable |
| MCP AtlasSource | 57.7% | — | Not comparable |
| ToolathlonSource | 42.9% | 49.5% | Step 3.7 Flash leads |
| τ²-bench resultsSource | 83.3% | 98.5% | Step 3.7 Flash leads |
| AA Agentic IndexSource | 30.2% | 21.5% | GPT-5.4 mini leads |
| APEX-Agents-AASource | 28.2% | 14.8% | GPT-5.4 mini leads |
| GDPval-AASource | 33.5% | 25.9% | GPT-5.4 mini leads |
| GDPval-AASource | 1171 | 1017 | GPT-5.4 mini leads |
| BrowseCompSource | — | 75.8% | Not comparable |
| DeepSearchQASource | — | 92.8% | Not comparable |
| Claw-EvalSource | — | 67.1% | Not comparable |
| HLE w/ toolsSource | — | 47.2% | Not comparable |
| Gert LabsSource | — | 51.57% | Not comparable |
Coding7 benchmarks
| Benchmark | GPT-5.4 mini | Step 3.7 Flash | Result |
|---|---|---|---|
| Vibe Code BenchSource | 47.97% | — | Not comparable |
| AA Coding IndexSource | 56.1% | 39.6% | GPT-5.4 mini leads |
| Terminal-Bench HardSource | 52.3% | 35.6% | GPT-5.4 mini leads |
| AA-SciCodeSource | 49.9% | 40.0% | GPT-5.4 mini leads |
| FrontierCodeSource | 27.0% | — | Not comparable |
| SWE-bench ProSource | — | 56.3% | Not comparable |
| Terminal-Bench 2.0Source | — | 59.5% | Not comparable |
Reasoning2 benchmarks
Knowledge9 benchmarks
| Benchmark | GPT-5.4 mini | Step 3.7 Flash | Result |
|---|---|---|---|
| GPQASource | 88% | — | Not comparable |
| HLESource | 41.5% | — | Not comparable |
| HLE w/o toolsSource | 28.2% | — | Not comparable |
| Artificial Analysis Intelligence IndexSource | 40.0% | 30.3% | GPT-5.4 mini leads |
| AA-GPQA DiamondSource | 87.5% | 80.9% | GPT-5.4 mini leads |
| AA-HLESource | 26.6% | 19.9% | GPT-5.4 mini leads |
| AA-Omniscience IndexSource | -18.7% | -37.5% | GPT-5.4 mini leads |
| AA-Omniscience AccuracySource | 37.5% | 25.4% | GPT-5.4 mini leads |
| AA-Omniscience Hallucination RateSource | 89.8% | 84.4% | Step 3.7 Flash leads |
Math2 benchmarks
Multimodal6 benchmarks
Inst. Following1 benchmarks
| Benchmark | GPT-5.4 mini | Step 3.7 Flash | Result |
|---|---|---|---|
| AA-IFBenchSource | 73.3% | 67.3% | GPT-5.4 mini leads |
Frequently Asked Questions (2)
Which is better, GPT-5.4 mini or Step 3.7 Flash?
GPT-5.4 mini is ahead on BenchLM's provisional leaderboard, 64 to 57. The biggest single separator in this matchup is Terminal-Bench 2.0, where the scores are 60% and 59.5%.
Which is better for agentic tasks, GPT-5.4 mini or Step 3.7 Flash?
Step 3.7 Flash has the edge for agentic tasks in this comparison, averaging 66.4 versus 65.7. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.
Related Comparisons
Explore More
The AI models change fast. We track them for you.
A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.