Model Comparison
Step 3.7 Flash
vs. Tri-21B-think Preview
Comparing 2 AI models · 7 benchmarks · StepFun, Trillion Labs
Recommended Pick
St Step 3.7 Flash 11 metric wins Strongest on: Throughput, Reasoning, Intelligence
Lowest Price
Tri-21B-think Preview
$0.00/1M input price
Best Reasoning
Step 3.7 Flash
47.8 reasoning score
Blends available reasoning benchmarks
Best for Coding
Step 3.7 Flash
37.1 coding index
Composite Indices
Higher is better; speed and price are normalized
Standard Benchmarks
Only benchmarks with data are shown
Differences That Matter
Price gap
Tri-21B-think Preview is ∞x cheaper on input tokens than Step 3.7 Flash.
Reasoning gap
Step 3.7 Flash leads Tri-21B-think Preview by 21.3 points on reasoning.
Coding gap
Step 3.7 Flash leads Tri-21B-think Preview by 29.7 points on coding.
Top-pick rationale
Step 3.7 Flash wins 11 measurable categories, including Throughput, Reasoning, Intelligence, Coding.
Response Face-Off
Run one prompt through the selected models and compare response quality with live speed and cost context.
Step 3.7 Flash
StepFun
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
Tri-21B-think Preview
Trillion Labs
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
Which answer was more useful?
Full Comparison
| Metric | Top Pick St Step 3.7 Flash | Tr Tri-21B-think Preview |
|---|---|---|
| Pricing per 1M tokens | ||
| Input Cost | $0.20/1M | $0.00/1M |
| Output Cost | $1.15/1M | $0.00/1M |
| Blended (3:1) | $0.44/1M | — |
| Specifications | ||
| Organization | StepFun | Trillion Labs |
| Release Date | May 29, 2026 | Feb 10, 2026 |
| Performance & Speed | ||
| Throughput | 365.0 tok/s | — |
| TTFT | 784ms | — |
| Latency | 6263ms | — |
| Composite Indices | ||
| Value Score | 100.0 | — |
| Reasoning Score | 47.8 | 26.5 |
| Intelligence | 42.6 | 20.0 |
| Coding | 37.1 | 7.4 |
| Standard Benchmarks | ||
| GPQA | 80.9% | 53.8% |
| HLE | 19.9% | 5.7% |
| SciCode | 40.0% | 17.8% |
| LCR | 63.7% | 14.7% |
| IFBench | 67.3% | 47.1% |
| TAU-bench v2 | 98.5% | 93.3% |
| TerminalBench Hard | 35.6% | 2.3% |
Key Takeaways
Tri-21B-think Preview offers the best value at $0.00/1M, making it ideal for high-volume applications and cost-conscious projects.
Step 3.7 Flash has the strongest reasoning profile with a 47.8 reasoning score, combining the available reasoning-heavy benchmarks.
Step 3.7 Flash reaches a 37.1 coding index, making it the top choice for software development and code generation tasks.
All models support context windows of ∞+ tokens, suitable for processing lengthy documents and maintaining extended conversations.
When to Choose Each Model
Step 3.7 Flash
- Complex reasoning tasks
- Research & analysis
- Code generation
- Software development
Tri-21B-think Preview
- Cost-sensitive applications
- High-volume processing
