Vascend
HorizonVascend IDEVascend CLICommunityBenchmarkContact
Sign inJoin Discord
HomeHorizonIDECLISign in
ChangelogIntelligenceCost per taskCoding Agent IndexCyber IndexDecision IndexBenchmarksSpeedLatencyPriceLeaderboard
Benchmark

Tri-21B-Think

  • Trillion Labs
  • Release · February 10, 2026
  • Open weights
  • Reasoning

Intelligence

Intelligence Index

#182 of 266
  1. 9.29.2Nova Premier
  2. 9.19.1Granite 4.2 3B
  3. 99gpt-oss-20b (High)
  4. 99Mistral Small 4 (Non-reasoning)
  5. 99K2-V2 (Medium)
  6. 99Tri-21B-Think
  7. 8.98.9Gemma 4 E4B (Reasoning)
  8. 8.98.9NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)
  9. 8.88.8MiniCPM5-1B (Reasoning)
  10. 8.88.8Sarvam 105B (High)
Data from Artificial Analysis

Evaluations

  • AA-Omniscience Accuracy12.3%#202
  • Humanity's Last Exam5.9%#191
  • AA-LCR15.7%#204
  • CritPt0.3%#117
  • GPQA Diamond60.1%#144
  • IFBench54.6%#53
  • τ²-Bench Telecom81%#35
  • Terminal-Bench Hard0.8%#115
Tri-21B-ThinkMedian
Data from Artificial Analysis

Nearby models

  • Nova PremierAmazon9.2
  • Granite 4.2 3BIBM9.1
  • gpt-oss-20b (High)OpenAI9
  • Mistral Small 4 (Non-reasoning)Mistral9
  • K2-V2 (Medium)Institute of Foundation Models9
  • Gemma 4 E4B (Reasoning)Google8.9
  • NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)NVIDIA8.9
  • MiniCPM5-1B (Reasoning)OpenBMB8.8
Data from Artificial Analysis
HorizonVascend IDEOlympusTalosCoursesCommunityBenchmark
DiscordGitHubXLet’s talkLegalPrivacyCookiesData useSecurity
© 2026 Vascend Technologies · VAT IT04911550616
ascend