Compare LLM Models
Select 2-5 models to compare side by side.
Select models to compare
0 / 5 selected
Top downloads
Select 2-5 models
| Attribute |
tokyotech-llm
|
|---|---|
| Company | tokyotech-llm |
| Release Date | 2026-02-20 |
| Parameters (B) | 20 |
| Architecture | MoE Transformer |
| Context Window | 128000 |
| Input Modalities | text |
| Open Source | Yes |
| License | apache-2.0 |
| Score | 0 |
| VRAM (GB) | 16 |
| Compute | Mixture-of-Experts Transformer (~21B total, 3.6B active per token); ~16 GB VRAM with MXFP4 quantization or ~42 GB for full BF16 weights; recommended inference via vLLM with reasoning effort set to medium |
| Benchmark: GPQA | 67.6 |
| Benchmark: HellaSwag | 83.1 |
| Benchmark: IFEval | 51.7 |
| Benchmark: LiveCodeBench | 66.5 |
| Benchmark: MATH-500 | 98.4 |
| Benchmark: MMLU-Pro | 73.8 |
| Pricing (Input) | — |
| Pricing (Output) | — |