Compare / Qwen3.8 27B vs Ternary Bonsai 2 27B
Qwen3.8 27B vs Ternary Bonsai 2 27B
Updated November 2026
Verdict
Qwen3.8 27B scores higher on the Small LM Index: 77.0 vs 74.6, although their likely ranges overlap, so the lead is not certain. On the 2 benchmarks both labs report, Qwen3.8 27B wins 2 and Ternary Bonsai 2 27B wins 0. Both need about 18.2 GB of memory at 4-bit quantization. Only Qwen3.8 27B has a hosted API on OpenRouter ($0.13 per million tokens); the other has to be self-hosted.
Pick Qwen3.8 27B if…
- you want the strongest model overall
- you mostly write code
- you need precise instruction following
- you want a hosted API rather than running it yourself
- you want the more widely used model (more quantized builds and community support)
Pick Ternary Bonsai 2 27B if…
No clear advantage in the data we have.
Specifications
| Qwen3.8 27B | Ternary Bonsai 2 27B | |
|---|---|---|
| Small LM Index | 77.0 (likely 76–79) | 74.6 (likely 72–77) |
| Lab | Alibaba (Qwen) · China | PrismML · United States |
| Released | 2026-08-14 | 2026-09-17 |
| Parameters | 27B (dense) | 27.4B (dense) |
| Memory at 4-bit | ~17.9 GB | ~18.2 GB |
| Context window | 262K tokens | 262K tokens |
| Input | text, image, video | text, image |
| Reasoning | Hybrid (thinking on/off) | Always thinks |
| License | Apache-2.0 | Apache-2.0 |
| API price / 1M tokens | $0.13 ($0.009 in, $0.49 out) | Self-host only |
| Hugging Face downloads (30 days) | 8M | — |
Benchmarks
Higher is better. Bold marks the winner where both labs report a score. ≈ marks the index model's estimate for a score that wasn't reported, ✓ an independent measurement and * a score measured by another lab.
| Benchmark | Qwen3.8 27B | Ternary Bonsai 2 27B |
|---|---|---|
| GPQA Diamond Knowledge | 89.2 | ≈ 85 |
| Humanity's Last Exam Reasoning | 30.8 | ≈ 24 |
| AIME Math | ≈ 96 | 95.8 |
| LiveCodeBench Coding | 90.3 | 90.1 |
| Terminal-Bench Agentic | 73.0 | — |
| IFBench Instruction following | 79.5 | 74.0 |
| MMMU-Pro Vision | — | 75.5 |
| MMLU Knowledge | — | 89.1 |
| IFEval Instruction following | — | 91.3 |
| BFCL Agentic | — | 74.9 |
| MATH-500 Math | — | 98.8 |
Scores are mostly reported by each lab with its own settings, so small gaps are noise. See the methodology. Open in the interactive comparison to add more models.
Questions
Which is better, Qwen3.8 27B or Ternary Bonsai 2 27B?
Qwen3.8 27B scores higher on the Small LM Index: 77.0 vs 74.6, although their likely ranges overlap, so the lead is not certain. On the 2 benchmarks both labs report, Qwen3.8 27B wins 2 and Ternary Bonsai 2 27B wins 0.
Which is cheaper to run?
Only Qwen3.8 27B has a hosted API on OpenRouter ($0.13 per million tokens); the other has to be self-hosted. Both need about 18.2 GB of memory at 4-bit quantization.
Which has the longer context window?
Both have a 262K-token context window.
Can I use Qwen3.8 27B and Ternary Bonsai 2 27B commercially?
Both are released under the Apache-2.0 license. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.