Trillion Labs · trillionlabs/tri-21b-think
| Provider | Trillion Labs |
|---|---|
| Family | tri |
| Type | llm-reasoning |
| Status | active |
| Released | 2026-02-18 |
| Updated | 2026-02-19 |
| Parameters | 20.7B |
| Open weights | yes |
| Relationship | Model |
|---|---|
| Is a finetune of | trillionlabs/Tri-21B |
Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.
| Device | Memory | Bandwidth | Best quality | Weights | tok/s | Fastest |
|---|---|---|---|---|---|---|
| Cerebras WSE-3 | 44.0 GB | 21000000.0 GB/s | fp8 | 20.73 GB | ~709270.5 | ~1418541.1 q4 |
| NVIDIA GB200 Grace Blackwell Superchip | 372.0 GB | 16000.0 GB/s | bf16 | 41.45 GB | ~270.2 | ~1080.8 q4 |
| AMD Instinct MI355X | 288.0 GB | 8000.0 GB/s | bf16 | 41.45 GB | ~135.1 | ~540.4 q4 |
| NVIDIA B200 | 180.0 GB | 7700.0 GB/s | bf16 | 41.45 GB | ~130.0 | ~520.1 q4 |
| Google TPU7x (Ironwood) | 192.0 GB | 7380.0 GB/s | bf16 | 41.45 GB | ~124.6 | ~498.5 q4 |
| AMD Instinct MI325X | 256.0 GB | 6000.0 GB/s | bf16 | 41.45 GB | ~101.3 | ~405.3 q4 |
| AMD Instinct MI300X | 192.0 GB | 5300.0 GB/s | bf16 | 41.45 GB | ~89.5 | ~358.0 q4 |
| NVIDIA H200 SXM | 141.0 GB | 4800.0 GB/s | bf16 | 41.45 GB | ~81.1 | ~324.2 q4 |
| NVIDIA H100 NVL | 94.0 GB | 3900.0 GB/s | bf16 | 41.45 GB | ~65.9 | ~263.4 q4 |
| NVIDIA H100 SXM | 80.0 GB | 3350.0 GB/s | bf16 | 41.45 GB | ~56.6 | ~226.3 q4 |
| AMD Instinct MI250X | 128.0 GB | 3200.0 GB/s | bf16 | 41.45 GB | ~54.0 | ~216.2 q4 |
| Google TPU v5p | 95.0 GB | 2765.0 GB/s | bf16 | 41.45 GB | ~46.7 | ~186.8 q4 |
| NVIDIA A100 80GB SXM | 80.0 GB | 2039.0 GB/s | bf16 | 41.45 GB | ~34.4 | ~137.7 q4 |
| NVIDIA H100 PCIe | 80.0 GB | 2000.0 GB/s | bf16 | 41.45 GB | ~33.8 | ~135.1 q4 |
| NVIDIA GeForce RTX 5090 | 32.0 GB | 1792.0 GB/s | fp8 | 20.73 GB | ~60.5 | ~121.0 q4 |
| Google TPU v6e (Trillium) | 32.0 GB | 1638.0 GB/s | fp8 | 20.73 GB | ~55.3 | ~110.6 q4 |
| AMD Instinct MI210 | 64.0 GB | 1600.0 GB/s | bf16 | 41.45 GB | ~27.0 | ~108.1 q4 |
| NVIDIA A100 40GB SXM | 40.0 GB | 1555.0 GB/s | fp8 | 20.73 GB | ~52.5 | ~105.0 q4 |
| Google TPU v4 | 32.0 GB | 1200.0 GB/s | fp8 | 20.73 GB | ~40.5 | ~81.1 q4 |
| NVIDIA GeForce RTX 4090 | 24.0 GB | 1008.0 GB/s | q6 | 15.54 GB | ~45.4 | ~68.1 q4 |
| AMD Radeon RX 7900 XTX | 24.0 GB | 960.0 GB/s | q6 | 15.54 GB | ~43.2 | ~64.8 q4 |
| NVIDIA GeForce RTX 5080 | 16.0 GB | 960.0 GB/s | q4 | 10.36 GB | ~64.8 | ~64.8 q4 |
| NVIDIA GeForce RTX 3090 | 24.0 GB | 936.0 GB/s | q6 | 15.54 GB | ~42.2 | ~63.2 q4 |
| NVIDIA L40S | 48.0 GB | 864.0 GB/s | fp8 | 20.73 GB | ~29.2 | ~58.4 q4 |
| Apple M3 Ultra | 512.0 GB | 819.0 GB/s | bf16 | 41.45 GB | ~13.8 | ~55.3 q4 |
| AMD Radeon RX 7900 XT | 20.0 GB | 800.0 GB/s | q5 | 12.95 GB | ~43.2 | ~54.0 q4 |
| Apple M2 Ultra | 192.0 GB | 800.0 GB/s | bf16 | 41.45 GB | ~13.5 | ~54.0 q4 |
| Google TPU v5e | 16.0 GB | 800.0 GB/s | q4 | 10.36 GB | ~54.0 | ~54.0 q4 |
| NVIDIA GeForce RTX 4080 SUPER | 16.0 GB | 736.0 GB/s | q4 | 10.36 GB | ~49.7 | ~49.7 q4 |
| NVIDIA GeForce RTX 4070 Ti SUPER | 16.0 GB | 672.0 GB/s | q4 | 10.36 GB | ~45.4 | ~45.4 q4 |
| AMD Radeon RX 9070 XT | 16.0 GB | 640.0 GB/s | q4 | 10.36 GB | ~43.2 | ~43.2 q4 |
| Apple M4 Max | 128.0 GB | 546.0 GB/s | bf16 | 41.45 GB | ~9.2 | ~36.9 q4 |
| Apple M1 Max | 64.0 GB | 400.0 GB/s | bf16 | 41.45 GB | ~6.8 | ~27.0 q4 |
| Apple M2 Max | 96.0 GB | 400.0 GB/s | bf16 | 41.45 GB | ~6.8 | ~27.0 q4 |
| Apple M3 Max | 128.0 GB | 400.0 GB/s | bf16 | 41.45 GB | ~6.8 | ~27.0 q4 |
| NVIDIA L4 | 24.0 GB | 300.0 GB/s | q6 | 15.54 GB | ~13.5 | ~20.3 q4 |
| NVIDIA GeForce RTX 4060 Ti 16GB | 16.0 GB | 288.0 GB/s | q4 | 10.36 GB | ~19.5 | ~19.5 q4 |
| Apple M4 Pro | 64.0 GB | 273.0 GB/s | bf16 | 41.45 GB | ~4.6 | ~18.4 q4 |
| Apple M4 | 32.0 GB | 120.0 GB/s | fp8 | 20.73 GB | ~4.1 | ~8.1 q4 |
Taken from the card's availability section.
| Platform | Id on that platform |
|---|---|
| Hugging Face | trillionlabs/Tri-21B-Think |
| Benchmark | Evaluated as | Score | Evidence date | Source kind | |
|---|---|---|---|---|---|
| aa_lcr | Tri-21B-Think | 15.67% | 2026-09-10 evaluated | independent evaluator | source |
| critpt | Tri-21B-Think | 0.29% | 2026-09-10 evaluated | independent evaluator | source |
| gpqa_diamond | Tri-21B-Think | 60.1% | 2026-09-10 evaluated | independent evaluator | source |
This card reports no benchmark scores yet.
| Benchmark | Catalogue standing | Score |
|---|