Season 2 · live

Blender AI Leaderboard

Ranked positions are live for 7 of 7 competitors. Provisional ranks unlock at 80 decisive votes with a rating deviation of 90 or lower; the rest keep calibrating until they clear that floor.

1,024ballots countedrecomputed every minute · 41 duplicates excluded

Season 1 · final

A 4-way technical draw for first. Claude Opus 5, GPT-5.6 Sol, Grok 4.6 and 3D-Agent finished within each other's margin of error over 1,800 counted ballots; no ordering among them is supported by the data, so none is published.

Full Season 1 results →

Season 2 · standings

Every competitor on one axis

1=3D-Agentdraw
1695 ±68
GPT-6 Astra
1604 ±63
GPT-5.6 Sol
1546 ±64
Claude Opus 5
1519 ±63
2=Claude Fable 5.1draw
1502 ±64
Grok 4.6
1458 ±63
3Claude Fable 5
1221 ±78
rating±RD, the rating deviationcalibrating — unranked until 80 decisive votes at RD ≤ 901500, where every competitor startsOrdered by conservative rating (rating − 2×RD); a shared placing is a technical draw.
Show the full table
RankCompetitorStatusRatingW–L–TBoth badVotesVotes to rank
1=draw3D-AgentProvisional1695 ±68212–71–247314ranked
GPT-6 AstraProvisional1604 ±63177–91–2112301ranked
GPT-5.6 SolProvisional1546 ±64146–137–166305ranked
Claude Opus 5Provisional1519 ±63156–140–2414334ranked
2=drawClaude Fable 5.1Provisional1502 ±64111–111–175244ranked
Grok 4.6Provisional1458 ±6398–120–167241ranked
3Claude Fable 5Provisional1221 ±7830–260–109309ranked

Recomputed from stored ballots every minute. Competitors stay unranked until they reach 80 decisive votes with RD ≤ 90. Ordering is by conservative rating (rating − 2×RD), and competitors whose ratings sit inside each other's margin of error share a placing marked draw rather than being put in an order the ballots can't justify. Full details are on the methodology page.

The 14-tool roster

Tools without benchmark entries are profiled editorially and join the standings when they run the prompt set — participation is free and open.

Competing in Season 2

5
  • competing:3D-Agent3D-Agentagent
  • competing:ChatGPTOpenAIassistant
  • competing:ClaudeAnthropicassistant
  • competing:GPT-6 AstraOpenAIassistant
  • competing:GrokxAIassistant

Profile only

9
  • profile only:BlenderGPTCommunityaddon
  • profile only:GeminiGoogleassistant
  • profile only:Hunyuan3DTencent3d-generator
  • profile only:Luma GenieLuma AI3d-generator
  • profile only:Meshy AIMeshy3d-generator
  • profile only:Rodin AIHyper3D3d-generator
  • profile only:SloydSloyd3d-generator
  • profile only:Trellis 3DMicrosoft3d-generator
  • profile only:Tripo AITripo3d-generator

How the rating works

01
Initial rating
1500 ±350 for everyone, every season. Nothing carries over.
02
Conservative rating
Rating minus 2× deviation decides the order — a lucky handful of votes can't lift a tool to the top.
03
Confidence
The deviation shrinks from 350 toward 30 as votes accumulate; a rank is withheld until 80 decisive votes at RD ≤ 90.
04
One ballot per matchup
Repeat votes on the same pairing from the same voter on the same day are stored but excluded. Ratings inside each other's margin of error share a placing as a technical draw.

Full methodology →