Tri-21B-Think
- Input
- —
- Output
- —
- Speed
- ~87tok/sest.
- Context
- 32Ktokens
Long-tail import from Artificial Analysis (2026-09-24); AA variant: Tri-21B-Think. Price = AA median host price.
Rankings
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.
- Hardware
- 1× B200 180GB
- BF16 weights
- Throughput
- ~5,640 tok/s
- many requests batched
- Price
- $0.16–1.46 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (int4, ~120 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- BF16 weights (20.73B params) + 25% KV-cache headroom ≈ 52 GB
- 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~5,640 output tok/s aggregate at batch 161 (bandwidth-bound)
- Blended 3:1 input:output; prefill ~25,642 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Writing & chat
IFBenchVerified
#67 of 13954.6%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #66LFM2.5-8B-A1B55.6%
- #67Tri-21B-Think54.6%
- #68Falcon-H1R-7B54.4%
artificialanalysis.ai · 2026-09-24
MMMLUEstimated
~#4 of 5 (estimated)~81.3%
- #1Gemini 3.1 Pro92.6%
- #2Gemma 4 31B88.4%
- #3Claude Haiku 4.583.0%
- ~#4Tri-21B-ThinkEstimated81.3%
- #4gpt-oss-120b81.3%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Estimated
~#45 of 57 (estimated)~1358 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #44GLM-4.6VFew votes1361
- ~#45Tri-21B-ThinkEstimated1358
- #45Trinity Large Thinking1350
No published score — estimated from related benchmarks
LMArena Text (Elo)Estimated
~#45 of 57 (estimated)~1378 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #44GLM-4.6VFew votes1378
- ~#45Tri-21B-ThinkEstimated1378
- #45Trinity Large Thinking1369
No published score — estimated from related benchmarks
EQ-Bench Creative Writing v3 (Elo)Estimated
~#19 of 26 (estimated)~1561 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #18Inkling1611
- ~#19Tri-21B-ThinkEstimated1561
- #19Gemini 3.5 Flash-Lite1559
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondVerified
#136 of 19260.1%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #135HyperCLOVA X SEED Think (32B)61.5%
- #136Tri-21B-Think60.1%
- #137Devstral 259.4%
artificialanalysis.ai · 2026-09-24
Humanity's Last ExamVerified
#146 of 1985.9%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #145Olmo 3 7B Think6.0%
- #146Tri-21B-Think5.9%
- #147LongCat Flash Lite5.8%
artificialanalysis.ai · 2026-09-24
AIME (latest)Estimated
~#23 of 34 (estimated)~51.2%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #22Phi-4-reasoning-plus78.0%
- ~#23Tri-21B-ThinkEstimated51.2%
- #23Mistral Large 338.0%
No published score — estimated from related benchmarks
AA Intelligence IndexEstimated
~#86 of 103 (estimated)~10.3
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #85Trinity Large Thinking10.8
- ~#86Tri-21B-ThinkEstimated10.3
- #86North Mini Code9.9
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Verified
#30 of 11981.0%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #30MiniCPM5-1B81.0%
- #30Tri-21B-Think81.0%
- #32Command A+80.7%
artificialanalysis.ai · 2026-09-24
OSWorld-VerifiedEstimated
~#12 of 15 (estimated)~62.4%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #11Muse Glimmer65.9%
- ~#12Tri-21B-ThinkEstimated62.4%
- #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~53.8%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#19 of 22 (estimated)~64.0%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #18Nemotron 3.5 Bolt 74B-A9B70.0%
- ~#19Tri-21B-ThinkEstimated64.0%
- #19Mistral Medium 3.548.6%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Estimated
~#47 of 92 (estimated)~910 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #46GPT-5.4 nano937
- ~#47Tri-21B-ThinkEstimated910
- #47Qwen3.6 35B A3B879
No published score — estimated from related benchmarks
AutomationBench-AAEstimated
~#57 of 85 (estimated)~3.5%
- #1Claude Fable 5.276.4%
- #2Muse Wonder Full71.3%
- #56Nemotron 3 Super 120B A12B3.8%
- ~#57Tri-21B-ThinkEstimated3.5%
- #57Claude Haiku 4.53.2%
No published score — estimated from related benchmarks
τ-Bench Banking (AA)Estimated
~#64 of 91 (estimated)~9.6%
- #1Muse Spark 1.350.5%
- #2GLM-5.350.3%
- #63Nemotron 3 Super 120B A12B10.3%
- ~#64Tri-21B-ThinkEstimated9.6%
- #64Claude Haiku 4.59.3%
No published score — estimated from related benchmarks