Benchmarks of Tri 21B Think11
BenchmarkByScoreRank
Overall1
Knowledge4
Science2
Coding1
Documents1
Legacy · saturated or no longer reported2
No benchmark matches
—per task?
self-host onlyprice
—output speed?
32Kcontext
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-6 Lunalow×2.4—$0.005Function calling only with none · Chat Completions+3
OpenQwen3.8 27Bxhigh×3.7—$1.15Memory (4-bit): 12 GB → 17 GB+5BetterGPT-5.6 Lunalow×2.3—$0.01
BetterMistral Small 4thinkingOpen+25%—$0.017Memory (4-bit): 12 GB → 72 GB
BetterGranite 4.2 8BthinkingOpen+23%—$0.024ModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Lunalow×2.4—$0.005Function calling only with none · Chat Completions+3GPT-5.6 Lunalow×2.3—$0.01Mistral Small 4thinkingOpen+25%—$0.017Memory (4-bit): 12 GB → 72 GBGranite 4.2 8BthinkingOpen+23%—$0.024
Claude Haiku 5.5low×3.3—$0.024Remove sampling parameters+5ModelScoreCostPer taskWhat changes
Qwen3.8 27BxhighOpen×3.7—$1.15Memory (4-bit): 12 GB → 17 GB+5IK2 Horizon MoVA 36B A4BthinkingOpen×2.8—$1 / $4Memory (4-bit): 12 GB → 22 GBQwen 3.5 27BthinkingOpen×2.5—$0.30 / $2.40Memory (4-bit): 12 GB → 17 GBAG9v3 39A5BOpen×2.4——Memory (4-bit): 12 GB → 23 GBQwen3.6 27BthinkingOpen×2.4—$0.62Memory (4-bit): 12 GB → 17 GBRun it yourself
estimateFits on✗ 8 GB (phone, laptop)✓ 16 GB · compressed✓ 24 GB GPU · 8-bit✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam5.9#254
AA Omniscience Index-62.6#273
CritPt (physics)0.3#151
AA Omniscience · Accuracy12.3#266
AA-LCR (long context) 1.115.7#250
GPQA Diamond60.1#228
Coding
Agents & Tool Use
τ²-bench · AA81#77
Writing & Language
IFBench (instructions)54.6#102
FAQ
How much does Tri 21B Think cost?›
There is no per-unit API price: self-host only.
Is Tri 21B Think free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Tri 21B Think, and what is it best at?›
#252 of 336 in Chat & Knowledge, #203 of 239 in Coding, #75 of 242 in Agents & Tool Use, #102 of 245 in Writing & Language.
Is Tri 21B Think better than Tri-21B-think Preview?›
On AA Intelligence Index 4.3.2, Tri 21B Think scores 9 vs 9.6 — about the same.
What is the context window of Tri 21B Think?›
32K tokens.
When was Tri 21B Think released? Is there a newer version?›
Released 2026-02-10. It is the current version in its line.
Is Tri 21B Think open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Tri 21B Think locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Tri 21B Think read images?›
No, it accepts text only.
What is Tri 21B Think?›
Tri 21B Think is an open-weights language model from Trillion Labs, released Feb 2026, the successor to Tri-21B-think Preview.
What are the best alternatives to Tri 21B Think?›
GPT-6 Luna (better), Qwen3.8 27B (open weights).