Model Comparison
V4 Pro (Non-reasoning)
vs. Qwen3 235B A22B 2507 Instruct
Comparing 2 AI models · 12 benchmarks · DeepSeek, Alibaba
Recommended Pick
De V4 Pro (Non-reasoning) 9 metric wins Strongest on: Throughput, TTFT, Latency
Best Value
Qwen3 235B A22B 2507 Instruct
100.0 value score
60.6 reasoning / $0.36/1M
Lowest Price
Qwen3 235B A22B 2507 Instruct
$0.20/1M input price
Best Reasoning
Qwen3 235B A22B 2507 Instruct
60.6 reasoning score
Blends available reasoning benchmarks
Best for Coding
V4 Pro (Non-reasoning)
38.4 coding index
Composite Indices
Higher is better; speed and price are normalized
Standard Benchmarks
Only benchmarks with data are shown
Differences That Matter
Best value
Qwen3 235B A22B 2507 Instruct has the strongest quality-to-price mix at 100.0 out of 100 value points.
Price gap
Qwen3 235B A22B 2507 Instruct is 2.2x cheaper on input tokens than V4 Pro (Non-reasoning).
Speed gap
V4 Pro (Non-reasoning) generates about 1.2x as many tokens per second as Qwen3 235B A22B 2507 Instruct.
Reasoning gap
Qwen3 235B A22B 2507 Instruct leads V4 Pro (Non-reasoning) by 21.0 points on reasoning.
Coding gap
V4 Pro (Non-reasoning) leads Qwen3 235B A22B 2507 Instruct by 16.3 points on coding.
Response Face-Off
Run one prompt through the selected models and compare response quality with live speed and cost context.
V4 Pro (Non-reasoning)
DeepSeek
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
Qwen3 235B A22B 2507 Instruct
Alibaba
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
Which answer was more useful?
Full Comparison
| Metric | Top Pick De V4 Pro (Non-reasoning) | Al Qwen3 235B A22B 2507 Instruct |
|---|---|---|
| Pricing per 1M tokens | ||
| Input Cost | $0.43/1M | $0.20/1M |
| Output Cost | $0.87/1M | $0.82/1M |
| Blended (3:1) | $0.54/1M | $0.36/1M |
| Specifications | ||
| Organization | DeepSeek | Alibaba |
| Release Date | Apr 24, 2026 | Jul 21, 2025 |
| Performance & Speed | ||
| Throughput | 80.8 tok/s | 67.1 tok/s |
| TTFT | 1179ms | 1248ms |
| Latency | 1179ms | 1248ms |
| Composite Indices | ||
| Value Score | 42.8 | 100.0 |
| Reasoning Score | 39.6 | 60.6 |
| Intelligence | 39.3 | 25.0 |
| Coding | 38.4 | 22.1 |
| Math | — | 71.7 |
| Standard Benchmarks | ||
| GPQA | 71.7% | 75.3% |
| MMLU Pro | — | 82.8% |
| HLE | 7.7% | 10.6% |
| LiveCodeBench | — | 52.4% |
| MATH 500 | — | 98.0% |
| AIME 2025 | — | 71.7% |
| AIME (Original) | — | 71.7% |
| SciCode | 42.4% | 36.0% |
| LCR | 45.0% | 31.2% |
| IFBench | 45.8% | 46.1% |
| TAU-bench v2 | 91.2% | 33.3% |
| TerminalBench Hard | 36.4% | 15.2% |
Key Takeaways
Qwen3 235B A22B 2507 Instruct offers the best value at $0.20/1M, making it ideal for high-volume applications and cost-conscious projects.
Qwen3 235B A22B 2507 Instruct has the strongest reasoning profile with a 60.6 reasoning score, combining the available reasoning-heavy benchmarks.
V4 Pro (Non-reasoning) reaches a 38.4 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
V4 Pro (Non-reasoning)
- Code generation
- Software development
Qwen3 235B A22B 2507 Instruct
- Cost-sensitive applications
- High-volume processing
- Complex reasoning tasks
- Research & analysis
