MODEL LEADERBOARDS
Leaderboard
Ranks come verbatim from each evaluator's own official board, synced on a schedule. The rulers differ (Arena is human-preference Elo, Terminal-Bench a task success rate, SWE-bench an issue resolve rate), so a score is only comparable within its own board — a cross-board composite would look comparable without being so. This page gives three things: who is on top, by how much, and when the score was measured.
- Boards
- 12
- Ranked entries
- 519
- Evaluators
- 6
Embodied AI
2 boards · 64 entriesRoboDojo Simulation Leaderboard
robodojo-benchmark.comranked by: Official composite scoreUnified embodied-policy eval: 42 sim tasks across generalization, precision, long-horizon, memory and open-ended dims.
| # | Model | Gen-Std | Gen-Rand | Precision | Long-Horizon | Memory | Open-Ended | Average |
|---|---|---|---|---|---|---|---|---|
| 1 | WALL-OSS-1.0-liteX Square Robot | 53.4 | 39.9 | 38.2 | 49.9 | 57.4 | 20.8 | 42.6 |
| 2 | VPP2RobotEra | 52.1 | 20.6 | 43 | 56.3 | 56 | 4.6 | 39.3 |
| 3 | Awomo-0.5Awomo | 37.7 | 22.9 | 39.7 | 40.4 | 42.3 | 24.1 | 35.3 |
| 4 | Simate-betaSimate | 40.5 | 29.6 | 34.4 | 57.8 | 33.3 | 9.1 | 34 |
| 5 | Rex-M1 PreviewFT Lab & VisIncept | 42.6 | 25.5 | 32.8 | 46.9 | 48.5 | 6.7 | 33.8 |
| 6 | HWM-0 PreviewHorizon Robotics | 37.6 | 16.4 | 37.6 | 46 | 46.9 | 6.6 | 32.8 |
| 7 | VPP2-PreviewRobotEra | 43.9 | 18.8 | 33.8 | 35.3 | 50.8 | 5.8 | 31.4 |
| 8 | InternW0-ΔShanghai AI Lab | 41 | 19.2 | 32 | 46.3 | 34.7 | 10.8 | 30.8 |
| 9 | Liber-0 PreviewLiberAI | 39.3 | 10.7 | 38.3 | 46 | 37.8 | 6.7 | 30.7 |
| 10 | Liber-0 LiteLiberAI | 39.8 | 10.9 | 36.6 | 45.5 | 35.7 | 3.1 | 29.2 |
| 11 | DM0.5Dexmal | 23.5 | 8.1 | 24.8 | 33.7 | 47.7 | 2.4 | 24.9 |
| 12 | ME-Brain-1.0Li Auto Inc. | 29.6 | 14 | 24.2 | 33 | 23.5 | 5.8 | 21.7 |
| 13 | GalaxeaVLA (G0.5)Galaxea AI | 27.7 | 9.2 | 28.3 | 44.1 | 8.6 | 1.7 | 20.2 |
| 14 | Xiaomi-Robotics-1Xiaomi Robotics | 35.7 | 11.4 | 26.7 | 38.4 | 7.8 | 3.9 | 20.1 |
| 15 | MachEmbodied-U0Li Auto Inc. | 25.3 | 6.9 | 24.2 | 36.7 | 8.2 | 3.8 | 17.8 |
| 16 | OpenWAM-αOpenWAM Team | 33.2 | 8.3 | 18.5 | 34.9 | 10.4 | 1.4 | 17.2 |
| 17 | EWAMEWAM Team | 24 | 6.2 | 22.2 | 30.2 | 7.4 | 1.1 | 15.2 |
| 18 | Meituan-Robotics-0Meituan Robotics | 19.3 | 8.2 | 16.8 | 29.6 | 10.1 | 4.5 | 15 |
| 19 | HWM-CogWAMHorizon Robotics | 23.5 | 2.1 | 20.5 | 27.4 | 8.2 | 2.9 | 14.3 |
| 20 | Hy-Embodied-0.5-VLATencent Robotics X | 22 | 1.6 | 13.8 | 25.7 | 13.4 | 0.7 | 13.1 |
| 21 | KinRTIIGroup | 20.9 | 7.1 | 15.7 | 26.4 | 4.8 | 4.2 | 13 |
| 22 | SimpleMemVLASimpleMemVLA Team | 10.4 | 2.3 | 7.4 | 14.6 | 33.7 | 0.9 | 12.6 |
| 23 | Spatial ForcingOpenHelix Robotics | 21.3 | 7 | 17.3 | 23.3 | 5.4 | 1.8 | 12.4 |
| 24 | Pi-05 | 20.9 | 5.8 | 12.4 | 23.5 | 5.9 | 2 | 11.4 |
| 25 | InternVLA-A1.5InternVLA Team | 16.8 | 3.9 | 15.2 | 23.8 | 4.9 | 1.4 | 11.2 |
RoboDojo Real-World Leaderboard
robodojo-benchmark.comranked by: Official composite scoreReal-robot embodied eval: 18 tasks on ARX X5, Piper and Piper X.
| # | Model | ARX X5 | Piper | Piper X | Average |
|---|---|---|---|---|---|
| 1 | SLA-PreviewHKUST(GZ) & OLA | 50.3 | 60.3 | 16.3 | 42.3 |
| 2 | InternW0-ΔShanghai AI Lab | 59.7 | 39.3 | 17.8 | 38.9 |
| 3 | OpenWAM-αOpenWAM Team | 39 | 46.7 | 27.2 | 37.6 |
| 4 | MWA-Phoenix-PreviewAnyverse & OpenPhoenix | 29 | 58 | 18.8 | 35.3 |
| 5 | Pi-05RoboDojo Team | 27.7 | 32 | 9.1 | 22.9 |
| 6 | InternVLA-A1RoboDojo Team | 10 | 23.2 | 2.7 | 12 |
| 7 | GalaxeaVLA (G0)RoboDojo Team | 3.4 | 22 | 1.7 | 9 |
| 8 | Xiaomi-Robotics-0RoboDojo Team | 15 | 7.6 | 1.2 | 7.9 |
| 9 | X-VLARoboDojo Team | 6.8 | 15.9 | 0.1 | 7.6 |
| 10 | GROOT-N1.7RoboDojo Team | 9.3 | 6.8 | 1.4 | 5.9 |
| 11 | Pi-0RoboDojo Team | 8.1 | 9 | 0.3 | 5.8 |
| 12 | StarVLARoboDojo Team | 0 | 12 | 0.5 | 4.1 |
| 13 | Spirit v1.5RoboDojo Team | 0 | 4 | 0.7 | 1.6 |
| 14 | DM0RoboDojo Team | 0 | 0 | 0 | 0 |
Large Language Models
2 boards · 100 entriesArtificial Analysis Intelligence Index
artificialanalysis.airanked by: Official intelligence index (higher is better)Artificial Analysis' own composite intelligence index: knowledge, reasoning and agentic evaluations folded into a single scale — the most widely cited cross-vendor measure of absolute capability. Different reasoning efforts of the same model are listed separately, as the source does.
| # | Model | Intelligence Index | Price in $/M | Price out $/M | Notes |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 (max)Anthropic | 64.56 | $8 | $40 | 1M · 2026-11-09 |
| 2 | Claude Fable 5.2 (xhigh)Anthropic | 62.22 | $8 | $40 | 1M · 2026-11-09 |
| 3 | Claude Fable 5.2 (high)Anthropic | 61.82 | $8 | $40 | 1M · 2026-11-09 |
| 4 | Gemini 4 Carbon (high)Google | 60.03 | $2 | $10 | 1M · 2026-10-28 |
| 5 | Claude Opus 5.5 (max)Anthropic | 59.11 | $4 | $20 | 1M · 2026-09-22 |
| 6 | Gemini 4 Carbon (medium)Google | 58.26 | $2 | $10 | 1M · 2026-10-28 |
| 7 | Claude Opus 5.5 (xhigh)Anthropic | 57.44 | $4 | $20 | 1M · 2026-09-22 |
| 8 | GPT-6 Astra (max)OpenAI | 56.85 | $10 | $50 | 1M · 2026-09-03 |
| 9 | Claude Fable 5.2 (medium)Anthropic | 56.41 | $8 | $40 | 1M · 2026-11-09 |
| 10 | GPT-6 Astra (xhigh)OpenAI | 55.31 | $10 | $50 | 1M · 2026-09-03 |
| 11 | GPT-6 Astra (high)OpenAI | 54.85 | $10 | $50 | 1M · 2026-09-03 |
| 12 | GPT-6.1 Sol (max)OpenAI | 54.61 | $2 | $10 | 1M · 2026-09-29 |
| 13 | Claude Sonnet 5.5 (max)Anthropic | 54.54 | $2 | $10 | 1M · 2026-09-28 |
| 14 | Gemini 4 Carbon (low)Google | 54.22 | $2 | $10 | 1M · 2026-10-28 |
| 15 | Claude Fable 5.2 (low)Anthropic | 53.92 | $8 | $40 | 1M · 2026-11-09 |
| 16 | Gemini 4 Hydrogen (xhigh)Google | 53.92 | $1.5 | $7.5 | 1M · 2026-11-12 |
| 17 | GPT-6.1 Sol (xhigh)OpenAI | 52.46 | $2 | $10 | 1M · 2026-09-29 |
| 18 | Claude Opus 5.5 (high)Anthropic | 51.84 | $4 | $20 | 1M · 2026-09-22 |
| 19 | GPT-6 Astra (medium)OpenAI | 51.55 | $10 | $50 | 1M · 2026-09-03 |
| 20 | GPT-6.1 Terra (max)OpenAI | 51.33 | $1 | $5 | 1M · 2026-10-26 |
| 21 | Gemini 4 Hydrogen (high)Google | 50.79 | $1.5 | $7.5 | 1M · 2026-11-12 |
| 22 | Gemini 4 Argon (high)Google | 50.41 | $2 | $10 | 1M · 2026-09-30 |
| 23 | GPT-6.1 Sol (high)OpenAI | 50.33 | $2 | $10 | 1M · 2026-09-29 |
| 24 | Muse Wonder Full (max)Meta | 50.08 | — | — | 1M · 2026-10-12 |
| 25 | Claude Sonnet 5.5 (xhigh)Anthropic | 49.92 | $2 | $10 | 1M · 2026-09-28 |
Arena Text Chatbot Leaderboard
arena.airanked by: Arena preference score (higher is better)The largest ongoing human-preference blind eval: Arena (formerly LMArena) ranks models by anonymous real-user battle votes.
| # | Model | Arena score | Votes |
|---|---|---|---|
| 1 | Gemini 4 Carbon (High)google | 1538±9.6 | 4,273 |
| 2 | Gemini 4 Argon (High)google | 1529±5.4 | 20,779 |
| 3 | Claude Opus 5.5 (High)anthropic | 1514±6.0 | 15,793 |
| 4 | Claude Fable 5.1 (Max)anthropic | 1512±5.5 | 20,511 |
| 5 | Claude Opus 5 (Max)anthropic | 1507±4.7 | 35,670 |
| 6 | Muse Wonder (Max)meta | 1503±6.2 | 12,696 |
| 7 | Claude Opus 5 (High)anthropic | 1502±3.6 | 73,632 |
| 8 | Gemini 4 Hydrogen (High)google | 1502±9.0 | 4,741 |
| 9 | Claude Opus 4.6 (High)anthropic | 1501±3.2 | 84,375 |
| 10 | Gemini 3.8 Flash (High)google | 1500±4.2 | 62,675 |
| 11 | Claude Opus 4.6anthropic | 1497±3.2 | 89,177 |
| 12 | Claude Fable 5 (High)anthropic | 1492±4.2 | 45,469 |
| 13 | Muse Spark 1.3 (Max)meta | 1492±5.0 | 26,806 |
| 14 | MiMo-V2.6-Proxiaomi | 1492±9.4 | 4,064 |
| 15 | Claude Opus 4.7 (High)anthropic | 1490±3.8 | 64,260 |
| 16 | Gemini 3.7 Flash (High)google | 1489±4.7 | 33,251 |
| 17 | Qwen3.8-Maxalibaba | 1485±4.5 | 36,936 |
| 18 | Muse Spark 1.2 (xHigh)meta | 1483±5.8 | 15,563 |
| 19 | Claude Opus 4.7anthropic | 1483±3.8 | 65,309 |
| 20 | Gemini 3.5 Flash (High)google | 1483±3.9 | 47,588 |
| 21 | Gemini 3.1 Pro Previewgoogle | 1480±2.9 | 132,181 |
| 22 | Gemini 3 Progoogle | 1480±3.8 | 41,803 |
| 23 | Gemini 3.6 Flash (High)google | 1479±4.4 | 35,362 |
| 24 | Muse Spark 1.1meta | 1479±4.1 | 45,519 |
| 25 | Gemini 3.5 Flash (Medium)google | 1478±4.1 | 46,329 |
Agents
2 boards · 100 entriesArena Agent Leaderboard
arena.airanked by: IPS score in percentage points (higher is better)Arena's aggregate Agent eval: an IPS score measures net improvement on real tool-use tasks, with official task cost.
| # | Model | IPS score (%) | Cost/task P50 ($) |
|---|---|---|---|
| 1 | Claude Fable 5.2 (Max)anthropic | 28.61%±3.7 | $9.02 |
| 2 | Claude Opus 5.5 (Max)anthropic | 18.19%±2.0 | $5.65 |
| 3 | Gemini 4 Carbon (High)google | 16.45%±0.7 | $1.35 |
| 4 | Claude Sonnet 5.5 (Max)anthropic | 15.29%±2.9 | $3.46 |
| 5 | Gemini 4 Hydrogen (High)google | 14.91%±1.9 | $1.53 |
| 6 | Claude Fable 5.1 (Max)anthropic | 14.1%±2.1 | $7.83 |
| 7 | GPT-6 Astra (Max)openai | 13.91%±2.0 | $5.28 |
| 8 | Claude Opus 5.5 (High)anthropic | 13.35%±1.0 | $2.12 |
| 9 | GPT-6.1 Sol (Max)openai | 11.78%±1.1 | $0.99 |
| 10 | GPT-6.1 Terra (Max)openai | 11.34%±1.2 | $0.87 |
| 11 | Gemini 4 Argon (High)google | 10%±0.9 | $0.7 |
| 12 | GPT-6.1 Sol (Medium)openai | 8.04%±1.0 | $0.37 |
| 13 | Muse Wonder Full (Max)meta | 8%±0.9 | — |
| 14 | Qwen4-Max (xHigh)alibaba | 7.46%±1.3 | $1.47 |
| 15 | Claude Sonnet 5.5 (High)anthropic | 7.18%±1.9 | $0.88 |
| 16 | GPT-6 Sol (Max)openai | 6.25%±1.6 | $1.85 |
| 17 | Claude Fable 5 (High)anthropic | 6.22%±1.2 | $3.33 |
| 18 | Claude Opus 5 (Max)anthropic | 6.18%±1.3 | $4.5 |
| 19 | Hy4tencent | 5.67%±1.3 | $0.57 |
| 20 | GPT-6.1 Terra (Medium)openai | 4.44%±0.6 | $0.16 |
| 21 | Kimi K3.1 Pro (Max)moonshot | 3.95%±1.1 | $1.16 |
| 22 | GPT-5.6 Sol (xHigh)openai | 3.72%±1.1 | $1.59 |
| 23 | Inkling 2 Smallthinky | 2.51%±1.3 | $0.26 |
| 24 | Grok 4.7 (xHigh)xai | 1.78%±1.5 | $3.2 |
| 25 | Muse Ember (Max)meta | 1.73%±1.5 | $0.02 |
Terminal-Bench 4.0
tbench.airanked by: Task success rate %Hard terminal-agent benchmark: success rate, cost and tokens on real shell/engineering tasks, maintained by the Laude Institute.
| # | Model | Accuracy % | pass@5 | Trials | Notes |
|---|---|---|---|---|---|
| 1 | Opus 5.5Anthropic · Claude Code | 64.9% | 78.8% | 330 | Claude Code · $4.7k · 2026-09-22 |
| 2 | Sonnet 5.5Anthropic · Claude Code | 61.8% | 77.3% | 330 | Claude Code · $7.3k · 2026-09-28 |
| 3 | Kimi K3.1 ProMoonshot AI · Claude Code | 60.9% | 84.9% | 330 | Claude Code · $1.1k · 2026-10-24 |
| 4 | GPT-6.1 SolOpenAI · Codex | 58.2% | 74.2% | 330 | Codex · $634.24 · 2026-09-29 |
| 4 | GPT-6 AstraOpenAI · Codex | 58.2% | 71.2% | 330 | Codex · $3.3k · 2026-09-03 |
| 6 | Fable 5.1Anthropic · Claude Code | 57.9% | 78.8% | 330 | Claude Code · $6.2k · 2026-09-01 |
| 6 | GPT-6 AstraOpenAI · Codex | 57.9% | 69.7% | 330 | Codex · $2.4k · 2026-09-03 |
| 6 | GPT-6 AstraOpenAI · Codex | 57.9% | 71.2% | 330 | Codex · $2.3k · 2026-09-03 |
| 6 | Fable 5.1Anthropic · Claude Code | 57.9% | 75.8% | 330 | Claude Code · $4.9k · 2026-09-01 |
| 10 | Gemini 4 ArgonGoogle · Antigravity CLI | 54.9% | 66.7% | 330 | Antigravity CLI · $1.1k · 2026-09-30 |
| 11 | Fable 5.1Anthropic · Claude Code | 54.6% | 75.8% | 330 | Claude Code · $4.0k · 2026-09-01 |
| 12 | GPT-6 AstraOpenAI · Codex | 54.2% | 66.7% | 330 | Codex · $1.9k · 2026-09-03 |
| 13 | Opus 5Anthropic · Claude Code | 53.9% | 69.7% | 330 | Claude Code · $6.1k · 2026-07-24 |
| 13 | Fable 5.1Anthropic · Claude Code | 53.9% | 71.2% | 330 | Claude Code · $2.8k · 2026-09-01 |
| 15 | Opus 5Anthropic · Claude Code | 51.8% | 69.7% | 330 | Claude Code · $6.0k · 2026-07-24 |
| 15 | GPT-6.1 TerraOpenAI · Codex | 51.8% | 63.6% | 330 | Codex · $874.44 · 2026-10-26 |
| 17 | GPT-6 AstraOpenAI · Codex | 50.6% | 63.6% | 330 | Codex · $1.6k · 2026-09-03 |
| 18 | Opus 5Anthropic · Claude Code | 50.3% | 74.2% | 330 | Claude Code · $4.7k · 2026-07-24 |
| 19 | GPT-6 SolOpenAI · Codex | 49.4% | 66.7% | 330 | Codex · $1.5k · 2026-09-22 |
| 20 | Muse WonderMeta · Muse Code | 47% | 71.2% | 330 | Muse Code · $2.5k · 2026-10-12 |
| 21 | Opus 5Anthropic · Claude Code | 44.9% | 69.7% | 330 | Claude Code · $3.2k · 2026-07-24 |
| 22 | Fable 5Anthropic · Claude Code | 44.6% | 68.2% | 330 | Claude Code · $7.3k · 2026-06-09 |
| 23 | Fable 5.1Anthropic · Claude Code | 43.3% | 65.1% | 330 | Claude Code · $2.4k · 2026-09-01 |
| 24 | GLM-5.3Z.ai · Claude Code | 41.8% | 57.6% | 330 | Claude Code · $2.7k · 2026-08-14 |
| 25 | Haiku 5.5Anthropic · Claude Code | 40% | 62.1% | 330 | Claude Code · $952.47 · 2026-10-07 |
Code
1 boards · 49 entriesSWE-bench Verified
swe-bench.comranked by: Issue resolve rate %Real GitHub-issue repair benchmark (OpenAI-verified subset): end-to-end agent resolve rate, the hardest public signal for coding models.
| # | Model | Resolved % | Cost $ | Notes |
|---|---|---|---|---|
| 1 | Sonar Foundation Agent + Claude 4.5 OpusAnthropic | 79.2% | — | 2025-12-05 |
| 2 | live-SWE-agent + Claude 4.5 OpusAnthropic | 79.2% | — | 2025-12-15 |
| 3 | TRAE + Doubao-Seed-CodeByteDance | 78.8% | — | 2025-09-28 |
| 4 | live-SWE-agent + Gemini 3 Pro PreviewGoogle DeepMind | 77.4% | — | 2025-11-20 |
| 5 | EPAM AI/Run Developer Agent + Claude 4 SonnetAnthropic | 76.8% | — | 2025-08-04 |
| 6 | Atlassian Rovo Dev + MultipleAnthropic | 76.8% | — | 2025-09-02 |
| 7 | mini-SWE-agent + Claude 4.5 OpusAnthropic | 76.8% | $376.95 | 2026-02-17 |
| 8 | ACoder + MultipleAnthropic | 76.4% | — | 2025-08-19 |
| 9 | mini-SWE-agent + Gemini 3 FlashGoogle DeepMind | 75.8% | $177.98 | 2026-02-17 |
| 10 | mini-SWE-agent + MiniMax M2.5Minimax | 75.8% | $36.64 | 2026-02-17 |
| 11 | Warp + MultipleOpenAI | 75.6% | — | 2025-09-01 |
| 12 | mini-SWE-agent + Claude 4.6 OpusAnthropic | 75.6% | $275.76 | 2026-02-17 |
| 13 | TRAE + MultipleAnthropic | 75.2% | — | 2025-06-12 |
| 14 | Harness AI + Claude Sonnet 4Anthropic | 74.8% | — | 2025-07-31 |
| 15 | Sonar Foundation Agent + Claude 4.5 SonnetAnthropic | 74.8% | — | 2025-11-03 |
| 16 | Lingxi-v1.5 + Claude 4 SonnetAnthropic | 74.6% | — | 2025-07-20 |
| 17 | JoyCode + MultipleAnthropic | 74.6% | — | 2025-09-15 |
| 18 | Refact.ai Agent + MultipleAnthropic | 74.4% | — | 2025-06-03 |
| 19 | Prometheus-v1.2.1 + GPT-5OpenAI | 74.4% | — | 2025-10-15 |
| 20 | mini-SWE-agent + Claude 4.5 OpusAnthropic | 74.4% | $360.62 | 2025-11-24 |
| 21 | mini-SWE-agent + Gemini 3 Pro PreviewGoogle DeepMind | 74.2% | $229.98 | 2025-11-18 |
| 22 | Salesforce AI Research SAGE + MultipleAnthropic | 73.8% | — | 2025-11-03 |
| 23 | Tools + Claude 4 OpusAnthropic | 73.2% | — | 2025-05-22 |
| 24 | Salesforce AI Research SAGE + MultipleAnthropic | 73% | — | 2025-10-21 |
| 25 | mini-SWE-agent + GLM 5Z-AI | 72.8% | $267.19 | 2026-02-17 |
Image Generation
2 boards · 100 entriesArena Text-to-Image Leaderboard
arena.airanked by: Arena preference score (higher is better)Arena text-to-image preference board: blind pairwise votes on identical prompts; Elo-style score and vote count.
| # | Model | Arena score | Votes |
|---|---|---|---|
| 1 | GPT-Image-2.5 Sunburstopenai | 1428±5.0 | 68,073 |
| 2 | GPT-Image-2.6 Sunburst (Medium)openai | 1410±9.3 | 6,547 |
| 3 | GPT-Image-2.5 Flareopenai | 1400±4.8 | 67,372 |
| 4 | MAI-Image-2.7microsoft-ai | 1387±11.9 | 3,082 |
| 5 | GPT-Image-2 (Medium)openai | 1382±3.8 | 110,696 |
| 6 | GPT-Image-2.6 Flare (Medium)openai | 1375±9.3 | 6,447 |
| 7 | MAI-Image-2.6microsoft-ai | 1331±4.9 | 42,837 |
| 8 | Gemini Nano Banana 2.1google | 1328±9.1 | 4,957 |
| 9 | Grok Imagine Image 2.0xai | 1318±7.5 | 9,894 |
| 10 | Reve 2.1reve | 1302±7.6 | 7,897 |
| 11 | Grok Imagine Image 2.0 (20260801)xai | 1287±5.0 | 38,904 |
| 12 | Reve 2.0reve | 1268±6.2 | 15,282 |
| 13 | Muse Imagemeta | 1266±4.4 | 62,120 |
| 14 | Gemini 3.1 Flash Image (nano-banana-2, Web Search)google | 1256±3.7 | 84,052 |
| 15 | Qwen-Image-3.0-Proalibaba | 1254±6.6 | 10,895 |
| 16 | MAI-Image-2.5microsoft-ai | 1253±4.2 | 58,997 |
| 17 | Seedream 5.0 Probytedance | 1249±3.7 | 112,977 |
| 18 | Gemini 3.1 Flash-Lite Image (nano-banana-2-lite)google | 1249±6.1 | 17,109 |
| 19 | Gemini 3 Pro Image (nano-banana-pro, 2K)google | 1246±3.0 | 184,617 |
| 20 | GPT-Image-1.5 (High Fidelity)openai | 1236±3.1 | 169,259 |
| 21 | Gemini 3 Pro Image Preview (nano-banana-pro)google | 1231±4.9 | 83,785 |
| 22 | Qwen-Image-2.1alibaba | 1227±6.3 | 17,794 |
| 23 | Ideogram 4.0 (Quality)ideogram | 1199±3.6 | 83,791 |
| 24 | MiniMax H3-Imageminimax | 1198±10.7 | 3,285 |
| 25 | Qwen-Image-2.0-Pro (2026-06-22)alibaba | 1190±5.9 | 12,648 |
Arena Image Edit Leaderboard
arena.airanked by: Arena preference score (higher is better)Arena image-edit preference board: blind rankings for instruction fidelity and edit quality.
| # | Model | Arena score | Votes |
|---|---|---|---|
| 1 | GPT-Image-2.5 Sunburstopenai | 1519±4.4 | 140,743 |
| 2 | GPT-Image-2.5 Flareopenai | 1484±4.2 | 137,820 |
| 3 | GPT-Image-2.6 Sunburst (Medium)openai | 1481±7.0 | 16,881 |
| 4 | MAI-Image-2.7microsoft-ai | 1475±11.0 | 3,007 |
| 5 | GPT-Image-2 (Medium)openai | 1462±2.9 | 371,930 |
| 6 | GPT-Image-2.6 Flare (Medium)openai | 1461±6.8 | 16,830 |
| 7 | Grok Imagine Image 2.0xai | 1437±5.8 | 33,908 |
| 8 | MAI-Image-2.6microsoft-ai | 1427±4.7 | 82,671 |
| 9 | Gemini Nano Banana 2.1google | 1427±6.0 | 12,041 |
| 10 | Grok Imagine Image 2.0 (20260801)xai | 1419±4.0 | 144,482 |
| 11 | Krea 3 Ultrakrea | 1412±11.6 | 2,625 |
| 12 | MAI-Image-2.5microsoft-ai | 1400±3.8 | 183,259 |
| 13 | Muse Imagemeta | 1400±3.6 | 211,112 |
| 14 | Seedream 5.0 Probytedance | 1392±3.1 | 358,335 |
| 15 | Grok Imagine Image (Quality, 20260519)xai | 1391±5.6 | 37,096 |
| 16 | Gemini 3 Pro Image (nano-banana-pro, 2K)google | 1391±2.5 | 682,032 |
| 17 | ChatGPT Image Latest (High Fidelity, 20251216)openai | 1387±2.7 | 657,235 |
| 18 | Gemini 3.1 Flash Image (nano-banana-2, Web Search)google | 1387±3.1 | 268,179 |
| 19 | Gemini 3 Pro Image Preview (nano-banana-pro)google | 1385±2.9 | 532,373 |
| 20 | Reve 2.1reve | 1373±6.1 | 19,872 |
| 21 | GPT-Image-1.5 (High Fidelity)openai | 1367±2.6 | 682,072 |
| 22 | Qwen-Image-2.1alibaba | 1363±6.1 | 37,959 |
| 23 | Reve 2.0reve | 1359±6.5 | 18,599 |
| 24 | MiniMax H3-Imageminimax | 1358±9.0 | 5,134 |
| 25 | Ideogram 4.5ideogram | 1347±4.1 | 120,488 |
Video Generation
2 boards · 100 entriesArena Text-to-Video Leaderboard
arena.airanked by: Arena preference score (higher is better)Arena text-to-video preference board: double-blind votes on motion quality, prompt adherence and aesthetics.
| # | Model | Arena score | Votes |
|---|---|---|---|
| 1 | Gemini Omni 1.2 Flashgoogle | 1540±15.6 | 1,714 |
| 2 | Gemini Omni 1.1 Flashgoogle | 1513±15.3 | 1,792 |
| 3 | gemini-omni-flashgoogle | 1509±8.4 | 46,707 |
| 4 | FLUX 3 Videobfl | 1491±17.0 | 1,314 |
| 5 | Grok Imagine Video 1.5xai | 1488±18.6 | 1,249 |
| 6 | Dreamina Seedance 2.0 (720p)bytedance | 1483±7.9 | 72,137 |
| 7 | Dreamina Seedance 2.5 (720p)bytedance | 1479±8.3 | 25,421 |
| 8 | Wan 3.0alibaba | 1476±11.6 | 3,060 |
| 9 | SkyReels V4skyreel | 1459±18.3 | 1,060 |
| 10 | Muse Videometa | 1457±14.6 | 2,225 |
| 11 | MiniMax H3minimax | 1456±8.1 | 29,257 |
| 12 | HappyHorse 1.0aorizon | 1429±12.5 | 22,896 |
| 13 | Sora 2 Proopenai | 1367±6.6 | 55,965 |
| 14 | Veo 3.1 (Audio)google | 1363±14.4 | 13,790 |
| 15 | Veo 3.1 (Audio, 1080p)google | 1363±10.2 | 25,636 |
| 16 | Veo 3.1 Fast (Audio)google | 1361±10.4 | 40,278 |
| 17 | Veo 3.1 Fast (Audio, 1080p)google | 1359±9.8 | 26,483 |
| 18 | Veo 3 Fast (Audio)google | 1347±11.3 | 25,277 |
| 19 | Sora 2openai | 1342±6.1 | 73,992 |
| 20 | Grok Imagine Video (720p)xai | 1341±6.2 | 190,109 |
| 21 | Veo 3 (Audio)google | 1340±12.6 | 19,032 |
| 22 | Wan 2.7alibaba | 1337±7.0 | 48,517 |
| 23 | Wan 2.6alibaba | 1324±7.1 | 75,753 |
| 24 | Seedance 1.5 Probytedance | 1256±6.7 | 77,992 |
| 25 | Veo 3google | 1254±11.2 | 15,017 |
Arena Image-to-Video Leaderboard
arena.airanked by: Arena preference score (higher is better)Arena image-to-video preference board: double-blind votes on first-frame fidelity and motion plausibility.
| # | Model | Arena score | Votes |
|---|---|---|---|
| 1 | Gemini Omni 1.2 Flashgoogle | 1503±11.8 | 2,666 |
| 2 | Dreamina Seedance 2.5 (720p)bytedance | 1494±8.0 | 42,724 |
| 3 | MiniMax H3minimax | 1493±5.6 | 121,481 |
| 4 | Gemini Omni 1.1 Flashgoogle | 1487±11.0 | 3,725 |
| 5 | Wan 3.0alibaba | 1482±10.8 | 3,437 |
| 6 | Dreamina Seedance 2.0 (720p)bytedance | 1480±7.0 | 148,231 |
| 7 | SkyReels V4skyreel | 1475±15.6 | 1,542 |
| 8 | gemini-omni-flashgoogle | 1463±5.5 | 136,190 |
| 9 | HiDream-O1-Video-1.0hidream | 1458±8.6 | 25,149 |
| 10 | Grok Imagine Video 1.5 (720p)xai | 1453±4.6 | 197,120 |
| 11 | FLUX 3 Video (20260811)bfl | 1451±5.9 | 84,298 |
| 12 | HappyHorse 1.0aorizon | 1443±10.2 | 73,331 |
| 13 | Wan 2.7alibaba | 1427±5.7 | 108,756 |
| 14 | Grok Imagine Video (720p)xai | 1411±4.2 | 624,829 |
| 15 | Veo 3.1 (Audio)google | 1399±10.5 | 25,246 |
| 16 | Veo 3.1 (Audio, 1080p)google | 1388±8.5 | 54,737 |
| 17 | Veo 3.1 Fast (Audio)google | 1386±8.8 | 102,572 |
| 18 | Grok Imagine Video (480p)xai | 1385±8.3 | 19,833 |
| 19 | Veo 3.1 Fast (Audio, 1080p)google | 1373±9.3 | 56,047 |
| 20 | Vidu Q3 Proshengshu | 1363±8.5 | 38,498 |
| 21 | Kling 3.0 Prokling | 1353±5.4 | 263,258 |
| 22 | Veo 3 (Audio)google | 1331±11.4 | 32,395 |
| 23 | Veo 3 Fast (Audio)google | 1327±9.4 | 41,228 |
| 24 | Wan 2.5 Previewalibaba | 1322±10.3 | 17,048 |
| 25 | Wan 2.6alibaba | 1309±6.6 | 141,910 |
3D Generation
1 boards · 6 entriesT3 Bench 3D Generation Leaderboard
t3bench.comranked by: GPT-4V judged averageOfficial text-to-3D benchmark: GPT-4V-judged averages on single-object, single-with-surroundings and multi-object sets.
| # | Model | Average | Single Obj. | Single w/ Surr. | Multi Obj. |
|---|---|---|---|---|---|
| 1 | ProlificDreamer | 43.3 | 49.4 | 44.8 | 35.8 |
| 2 | Magic3D | 32.7 | 37 | 35.4 | 25.7 |
| 3 | LatentNeRF | 28.1 | 33.1 | 30.6 | 20.6 |
| 4 | Fantasia3D | 24 | 26.4 | 27 | 18.5 |
| 5 | DreamFusion | 21.7 | 24.4 | 24.6 | 16.1 |
| 6 | SJC | 18.7 | 24.7 | 19.8 | 11.7 |
As an Amazon Associate, we earn from qualifying purchases.