Model Comparison
Qwen3 8B (Non-reasoning)
vs. 105B (high)
Comparing 2 AI models · 12 benchmarks · Alibaba, Sarvam
Recommended Pick
Sa 105B (high) 14 metric wins Strongest on: Value, Input price, Output price
Best Value
105B (high)
100.0 value score
34.0 reasoning / $0.07/1M
Lowest Price
105B (high)
$0.04/1M input price
Best Reasoning
105B (high)
34.0 reasoning score
Blends available reasoning benchmarks
Best for Coding
105B (high)
9.8 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
105B (high) has the strongest quality-to-price mix at 100.0 out of 100 value points.
Price gap
105B (high) is 4.3x cheaper on input tokens than Qwen3 8B (Non-reasoning).
Speed gap
105B (high) generates about 1.4x as many tokens per second as Qwen3 8B (Non-reasoning).
Reasoning gap
105B (high) leads Qwen3 8B (Non-reasoning) by 3.4 points on reasoning.
Coding gap
105B (high) leads Qwen3 8B (Non-reasoning) by 2.7 points on coding.
Response Face-Off
Run one prompt through the selected models and compare response quality with live speed and cost context.
Qwen3 8B (Non-reasoning)
Alibaba
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
105B (high)
Sarvam
TTFT
—
Time
—
tok/s
—
Tokens
—
Cost
—
Which answer was more useful?
Full Comparison
| Metric | Al Qwen3 8B (Non-reasoning) | Top Pick Sa 105B (high) |
|---|---|---|
| Pricing per 1M tokens | ||
| Input Cost | $0.18/1M | $0.04/1M |
| Output Cost | $0.20/1M | $0.17/1M |
| Blended (3:1) | $0.18/1M | $0.07/1M |
| Specifications | ||
| Organization | Alibaba | Sarvam |
| Release Date | Apr 28, 2025 | Mar 6, 2026 |
| Performance & Speed | ||
| Throughput | 64.5 tok/s | 91.9 tok/s |
| TTFT | 1293ms | 1222ms |
| Latency | 1293ms | 22983ms |
| Composite Indices | ||
| Value Score | 36.0 | 100.0 |
| Reasoning Score | 30.6 | 34.0 |
| Intelligence | 10.6 | 18.2 |
| Coding | 7.1 | 9.8 |
| Math | 24.3 | — |
| Standard Benchmarks | ||
| GPQA | 45.2% | 73.8% |
| MMLU Pro | 64.3% | — |
| HLE | 2.8% | 10.1% |
| LiveCodeBench | 20.2% | — |
| MATH 500 | 82.8% | — |
| AIME 2025 | 24.3% | — |
| AIME (Original) | 24.3% | — |
| SciCode | 16.8% | 26.4% |
| LCR | 0.0% | 0.0% |
| IFBench | 28.6% | 34.4% |
| TAU-bench v2 | 24.9% | 46.8% |
| TerminalBench Hard | 2.3% | 1.5% |
Key Takeaways
105B (high) offers the best value at $0.04/1M, making it ideal for high-volume applications and cost-conscious projects.
105B (high) has the strongest reasoning profile with a 34.0 reasoning score, combining the available reasoning-heavy benchmarks.
105B (high) reaches a 9.8 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
Qwen3 8B (Non-reasoning)
- General-purpose AI
- Versatile applications
105B (high)
- Cost-sensitive applications
- High-volume processing
- Complex reasoning tasks
- Research & analysis
- Code generation
- Software development
