Benchmark Almanac

UnityBench-Mini3D

GPT-5.6 Sol 72.5% leads since October 2026
Not in model cards

Agents modify existing Unity projects. EditMode and PlayMode tests check the saved scene, component settings, and runtime behavior, with partial credit for completed milestones.

Unity publishes 36 configurations covering 21 models on 77 tasks. Its repository shows reference scenes and lists task projects and grading software as forthcoming.

Released
October 2026
Kind
Building whole projects
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text, Code → Actions, Code
Interaction
Agentic
Environment
Unity
Refresh
One fixed set
Run by
A host
In views
Agents
Strongest model tested
Claude Opus 5 #11 on the capability index
Last model added
August 2026
Scores refreshed
November 20th, 2026
21
Entries on its own leaderboard, 20 models we track

GPT-5.6 Sol leads with 72.5% among models released by August 2026 GPT-5.6 Sol leads with 72.5% among models released by August 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.6 Sol 72.5%
  2. 2 Claude Opus 5 71.8%
  3. 3 Grok 4.5 71.2%
  4. 4 Grok 4.6 70.6%
  5. 5 DeepSeek-V4-Pro 70.5%
  6. 6 GPT-5.6 Terra 68%
  7. 7 Qwen 3.8 Max 66.7%
  8. 8 GPT-5.6 Luna 66.4%
  9. 9 Kimi K3 66.3%
  10. 10 GPT-5.5 66.2%
  11. 11 DeepSeek-V4-Flash 65%
  12. 12 GLM-5.2 62.5%
  13. 13 Claude Sonnet 5 54%
  14. 14 MiniMax-M3 52.3%
  15. 15 Hunyuan 3 51%
  16. 16 Qwen3.7-Max 47.8%
  17. 17 Kimi K2.7 Code 41.5%
  18. 18 Qwen3.7-Plus 38.9%
  19. 19 Gemini 3.7 Flash 31.2%
  20. 20 Gemini 3.1 Pro 25.2%
  21. 21 Seed2.1 Pro 24.4%

Source: Official UnityBench-Mini3D leaderboard

#EntryScoreSettingDateSource
1 GPT-5.6 Sol 72.5% Codex Jul 2026 Official leaderboard
2 Claude Opus 5 71.8% Claude Code Jul 2026 Official leaderboard
3 Grok 4.5 71.2% Grok Build Jul 2026 Official leaderboard
4 Grok 4.6 70.6% Grok Build Aug 2026 Official leaderboard
5 DeepSeek-V4-Pro 70.5% DS Harness Apr 2026 Official leaderboard
6 GPT-5.6 Terra 68% Codex Jul 2026 Official leaderboard
7 Qwen 3.8 Max 66.7% Codely Aug 2026 Official leaderboard
8 GPT-5.6 Luna 66.4% Codex Jul 2026 Official leaderboard
9 Kimi K3 66.3% Kimi Code Jul 2026 Official leaderboard
10 GPT-5.5 66.2% Codex Apr 2026 Official leaderboard
11 DeepSeek-V4-Flash 65% DS Harness Apr 2026 Official leaderboard
12 GLM-5.2 62.5% Codely Jun 2026 Official leaderboard
13 Claude Sonnet 5 54% Claude Code Jun 2026 Official leaderboard
14 MiniMax-M3 52.3% Codely Jun 2026 Official leaderboard
15 Hunyuan 3 51% Codely Official leaderboard
16 Qwen3.7-Max 47.8% Qwen Code May 2026 Official leaderboard
17 Kimi K2.7 Code 41.5% Kimi Code Jun 2026 Official leaderboard
18 Qwen3.7-Plus 38.9% Qwen Code Jun 2026 Official leaderboard
19 Gemini 3.7 Flash 31.2% Antigravity Aug 2026 Official leaderboard
20 Gemini 3.1 Pro 25.2% Antigravity Feb 2026 Official leaderboard
21 Seed2.1 Pro 24.4% Codely Jun 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.6 Sol 72.5% Codex Jul 2026 Official leaderboard
2 Claude Opus 5 71.8% Claude Code Jul 2026 Official leaderboard
3 Grok 4.5 71.2% Grok Build Jul 2026 Official leaderboard
4 Grok 4.6 70.6% Grok Build Aug 2026 Official leaderboard
5 DeepSeek-V4-Pro 70.5% DS Harness Apr 2026 Official leaderboard
6 GPT-5.6 Terra 68% Codex Jul 2026 Official leaderboard
7 Qwen 3.8 Max 66.7% Codely Aug 2026 Official leaderboard
8 GPT-5.6 Luna 66.4% Codex Jul 2026 Official leaderboard
9 Kimi K3 66.3% Kimi Code Jul 2026 Official leaderboard
10 GPT-5.5 66.2% Codex Apr 2026 Official leaderboard
11 DeepSeek-V4-Flash 65% DS Harness Apr 2026 Official leaderboard
12 GLM-5.2 62.5% Codely Jun 2026 Official leaderboard
13 Claude Sonnet 5 54% Claude Code Jun 2026 Official leaderboard
14 MiniMax-M3 52.3% Codely Jun 2026 Official leaderboard
15 Qwen3.7-Max 47.8% Qwen Code May 2026 Official leaderboard
16 Kimi K2.7 Code 41.5% Kimi Code Jun 2026 Official leaderboard
17 Qwen3.7-Plus 38.9% Qwen Code Jun 2026 Official leaderboard
18 Gemini 3.7 Flash 31.2% Antigravity Aug 2026 Official leaderboard
19 Gemini 3.1 Pro 25.2% Antigravity Feb 2026 Official leaderboard
20 Seed2.1 Pro 24.4% Codely Jun 2026 Official leaderboard

Rankings by part

We read 10 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as User interface, Animation and Editor operations. Claude Opus 5 leads 3 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

User interface

20 models we track have a score here.

  1. 1 Qwen 3.8 Max ↑6 74.3%
  2. 2 GPT-5.6 Terra ↑4 72.8%
  3. 3 Claude Opus 5 ↓1 71.1%
  4. 4 GPT-5.5 ↑6 67.7%
  5. 5 Grok 4.6 ↓1 66.1%
  6. 6 GPT-5.6 Sol ↓5 65.8%
  7. 7 Kimi K3 ↑2 65.6%
  8. 8 DeepSeek-V4-Pro ↓3 64.4%
  9. 9 GLM-5.2 ↑3 63.9%
  10. 10 DeepSeek-V4-Flash ↑1 61.3%

The best score, 72.5%, costs $25 a task

Runs on the cost frontier score higher than any run that's cheaper. Grok 4.5 comes within 1.3 points of the best for $7.99 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Qwen xAI OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $1.00 $2.00 $5.00 $10 $20 $50 $100 Cost per task, log scale GPT-5.6 Sol Grok 4.5 GPT-5.6 Luna Qwen3.7-Plus 0% 20% 40% 60% 80% 100% $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol Grok 4.5

Source: Official UnityBench-Mini3D leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.6 Sol Codex 72.5% $25
Claude Opus 5 Claude Code 71.8% $50
Grok 4.5 Grok Build 71.2% $7.99
Grok 4.6 Grok Build 70.6% $12
DeepSeek-V4-Pro DS Harness 70.5% $10
Claude Opus 5 Codely 68.8% $50
GPT-5.6 Terra Codex 68% $14
Qwen 3.8 Max Codely 66.7% $13
GPT-5.6 Luna Codex 66.4% $1.94
Kimi K3 Kimi Code 66.3% $11
Kimi K3 Codely 66.2% $17
GPT-5.5 Codex 66.2% $23
DeepSeek-V4-Flash DS Harness 65% $4.98
Grok 4.6 Codely 63.9% $7.08
DeepSeek-V4-Pro Codely 63.1% $8.25
GLM-5.2 Codely 62.5% $12
GPT-5.6 Sol Codely 62.3% $19
Qwen 3.8 Max Qwen Code 62.3% $13
Kimi K3 OpenCode 62% $10
DeepSeek-V4-Flash Codely 61.7% $4.03
GPT-5.6 Sol OpenCode 61.3% $20
Claude Sonnet 5 Claude Code 54% $17
MiniMax-M3 Codely 52.3% $5.28
GLM-5.2 OpenCode 52% $7.49
Qwen3.7-Max Qwen Code 47.8% $5.74
Kimi K2.7 Code Kimi Code 41.5% $4.29
Qwen3.7-Plus Qwen Code 38.9% $1.16
DeepSeek-V4-Flash OpenCode 38.7% $1.09
Gemini 3.7 Flash Antigravity 31.2% $6.48
Gemini 3.1 Pro Antigravity 25.2% $1.81
Seed2.1 Pro Codely 24.4% $1.44
GPT-5.6 Sol UI-TARS 23.1% $3.74
GPT-5.6 Sol Agent S 22.2%
Kimi K3 UI-TARS 8.77% $2.02
Kimi K3 Agent S 5.04%

The best score, 72.5%, costs $25 a task

Runs on the cost frontier score higher than any run that's cheaper. Grok 4.5 comes within 1.3 points of the best for $7.99 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Qwen xAI OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $1.00 $2.00 $5.00 $10 $20 $50 $100 Cost per task, log scale GPT-5.6 Sol Grok 4.5 GPT-5.6 Luna Qwen3.7-Plus 0% 20% 40% 60% 80% 100% $1.00 $10 $100 Cost per task, log scale GPT-5.6 Sol Grok 4.5

Source: Official UnityBench-Mini3D leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.6 Sol Codex 72.5% $25
Claude Opus 5 Claude Code 71.8% $50
Grok 4.5 Grok Build 71.2% $7.99
Grok 4.6 Grok Build 70.6% $12
DeepSeek-V4-Pro DS Harness 70.5% $10
Claude Opus 5 Codely 68.8% $50
GPT-5.6 Terra Codex 68% $14
Qwen 3.8 Max Codely 66.7% $13
GPT-5.6 Luna Codex 66.4% $1.94
Kimi K3 Kimi Code 66.3% $11
Kimi K3 Codely 66.2% $17
GPT-5.5 Codex 66.2% $23
DeepSeek-V4-Flash DS Harness 65% $4.98
Grok 4.6 Codely 63.9% $7.08
DeepSeek-V4-Pro Codely 63.1% $8.25
GLM-5.2 Codely 62.5% $12
GPT-5.6 Sol Codely 62.3% $19
Qwen 3.8 Max Qwen Code 62.3% $13
Kimi K3 OpenCode 62% $10
DeepSeek-V4-Flash Codely 61.7% $4.03
GPT-5.6 Sol OpenCode 61.3% $20
Claude Sonnet 5 Claude Code 54% $17
MiniMax-M3 Codely 52.3% $5.28
GLM-5.2 OpenCode 52% $7.49
Qwen3.7-Max Qwen Code 47.8% $5.74
Kimi K2.7 Code Kimi Code 41.5% $4.29
Qwen3.7-Plus Qwen Code 38.9% $1.16
DeepSeek-V4-Flash OpenCode 38.7% $1.09
Gemini 3.7 Flash Antigravity 31.2% $6.48
Gemini 3.1 Pro Antigravity 25.2% $1.81
Seed2.1 Pro Codely 24.4% $1.44
GPT-5.6 Sol UI-TARS 23.1% $3.74
GPT-5.6 Sol Agent S 22.2%
Kimi K3 UI-TARS 8.77% $2.02
Kimi K3 Agent S 5.04%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard Not tested since Oct 2026
GPT-5.6 Sol 72.5%

Source: Official UnityBench-Mini3D leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol DeepSeek-V4-Pro GPT-5.5 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol GPT-5.5

Source: Official UnityBench-Mini3D leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Coding

All 523 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (21 entries).
  • Pass Models scored 20 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from August 2026, for Gemini 3.7 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
  • Pass Headroom The best is 72.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on unitybench.github.io: 35 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check