Benchmarks of Tri-21B-think Preview11
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×2.2—$0.005Function calling only with none · Chat Completions+3
OpenQwen3.8 27Bxhigh×3.5—$1.15Memory (4-bit): 12 GB → 17 GB+5BetterGPT-5.6 Lunalow×2.2—$0.01
BetterClaude Haiku 5.5low×3.1—$0.024Remove sampling parameters+5
BetterLing 3.0 FlashthinkingOpen×2.1—$0.028Memory (4-bit): 12 GB → 76 GBModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Lunalow×2.2—$0.005Function calling only with none · Chat Completions+3GPT-5.6 Lunalow×2.2—$0.01Claude Haiku 5.5low×3.1—$0.024Remove sampling parameters+5Ling 3.0 FlashthinkingOpen×2.1—$0.028Memory (4-bit): 12 GB → 76 GB
MiMo v2.5thinkingOpen×2.3—$0.042Memory (4-bit): 12 GB → 186 GBModelScoreCostPer taskWhat changes
Qwen3.8 27BxhighOpen×3.5—$1.15Memory (4-bit): 12 GB → 17 GB+5IK2 Horizon MoVA 36B A4BthinkingOpen×2.6—$1 / $4Memory (4-bit): 12 GB → 22 GBQwen 3.5 27BthinkingOpen×2.4—$0.30 / $2.40Memory (4-bit): 12 GB → 17 GBAG9v3 39A5BOpen×2.3——Memory (4-bit): 12 GB → 23 GBQwen3.6 27BthinkingOpen×2.2—$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 Exam5.8#257
AA Omniscience Index-62.3#271
CritPt (physics)0#182
AA Intelligence Index 4.3.2≈9.6#217
AA Omniscience · Accuracy10.1#281
AA-LCR (long context) 1.121.7#236
GPQA Diamond53.8#249
Coding
Agents & Tool Use
τ²-bench · AA93.3#29
Writing & Language
IFBench (instructions)47.1#125
FAQ
How much does Tri-21B-think Preview cost?›
There is no per-unit API price: self-host only.
Is Tri-21B-think Preview free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Tri-21B-think Preview, and what is it best at?›
#255 of 336 in Chat & Knowledge, #181 of 239 in Coding, #29 of 242 in Agents & Tool Use, #124 of 245 in Writing & Language. Strongest at: Tool calling.
Is Tri-21B-think Preview better than Qwen3 Next 80B A3B Instruct?›
On AA Intelligence Index 4.3.2, Tri-21B-think Preview scores 9.6 vs 9.6 — about the same.
What is the context window of Tri-21B-think Preview?›
32K tokens.
When was Tri-21B-think Preview released? Is there a newer version?›
Released 2026-02-10. It has been superseded by Tri 21B Think.
Is Tri-21B-think Preview open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Tri-21B-think Preview locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Tri-21B-think Preview read images?›
No, it accepts text only.
What is Tri-21B-think Preview?›
Tri-21B-think Preview is an open-weights language model from Trillion Labs, released Feb 2026.
What are the best alternatives to Tri-21B-think Preview?›
GPT-6 Luna (better), Qwen3.8 27B (open weights).