Skip to content
Model Pareto

Tri-21B-Think

Trillion LabsOpen weightsartificialanalysis.ai ↗
Compare with nearest rivalsCompare with…
Input
—
Output
—
Speed
~90tok/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

MMMLUEstimated
~#5 of 5 (estimated)~81.2%
  1. #1Gemini 3.1 Pro92.6%
  2. #2Gemma 4 31B88.4%
  3. #4gpt-oss-120b81.3%
  4. ~#5Tri-21B-ThinkEstimated81.2%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Estimated
~#42 of 54 (estimated)~1355 Elo
  1. #1Claude Opus 5.51501
  2. #2Claude Fable 5.1Few votes1496
  3. #41GLM-4.6VFew votes1361
  4. ~#42Tri-21B-ThinkEstimated1355
  5. #42Trinity Large Thinking1353
No published score — estimated from related benchmarks
LMArena Text (Elo)Estimated
~#42 of 54 (estimated)~1375 Elo
  1. #1Claude Opus 5.51509
  2. #2Claude Fable 5.11501
  3. #41GLM-4.6VFew votes1378
  4. ~#42Tri-21B-ThinkEstimated1375
  5. #42Trinity Large Thinking1369
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#14 of 25 (estimated)~49.9%
  1. #1Inkling97.1%
  2. #2Muse Glimmer94.7%
  3. #13Phi-4-reasoning-plus78.0%
  4. ~#14Tri-21B-ThinkEstimated49.9%
  5. #14Mistral Large 338.0%
No published score — estimated from related benchmarks

Agents

OSWorld-VerifiedEstimated
~#11 of 12 (estimated)~65.5%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #10Muse Glimmer65.9%
  4. ~#11Tri-21B-ThinkEstimated65.5%
  5. #11Claude Haiku 4.550.7%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~53.8%
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. ~#3Tri-21B-ThinkEstimated53.8%
  4. #3GPT-5.6 Terra50.2%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#15 of 18 (estimated)~61.0%
  1. #1Atria Dawn Preview92.5%
  2. #2GPT-6 Astra91.5%
  3. #14Inkling77.1%
  4. ~#15Tri-21B-ThinkEstimated61.0%
  5. #15Mistral Medium 3.548.6%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Estimated
~#42 of 87 (estimated)~915 Elo
  1. #1Claude Opus 5.51846
  2. #2Claude Fable 5.11735
  3. #41GPT-5.4 nano937
  4. ~#42Tri-21B-ThinkEstimated915
  5. #42Qwen3.6 35B A3B879
No published score — estimated from related benchmarks
τ-Bench Banking (AA)Estimated
~#64 of 89 (estimated)~9.2%
  1. #1Muse Spark 1.350.5%
  2. #2GLM-5.350.3%
  3. #62Qwen3.6 35B A3B9.3%
  4. ~#64Tri-21B-ThinkEstimated9.2%
  5. #64Nemotron 3.5 Lightning8.9%
No published score — estimated from related benchmarks