AI Model Leaderboard
Which AI model actually performs best? Not marketing claims. Not synthetic benchmarks. Every text and image generated on OneAIWorld is scored by an independent AI reviewer — these are the live results.
Run Your Own ComparisonLast 90 days of generations. Text models need 5+ and image models 3+ reviewed generations to be ranked — small samples can't distort the board.
Text Models
Scored 0–10 on accuracy, relevance, coherence, creativity, and language quality, plus the reviewer's overall rating — that's the ranking. Strongest At shows where each model beats the field average, so you can pick a model for your kind of task.
| Model | Overall | Accuracy | Relevance | Coherence | Creativity | Language Quality | Avg Speed | Reviews | Strongest At | |
|---|---|---|---|---|---|---|---|---|---|---|
| 🥇 | gemma-4-26b-moe | 9.44 | 9.5 | 10.0 | 9.57 | 8.36 | 9.79 | 41.4s | 14 | Creativity |
| 🥈 | gemma-4-31b | 9.38 | 9.42 | 10.0 | 9.33 | 8.33 | 9.83 | 51.6s | 12 | Creativity |
| 🥉 | claude-haiku-4-5 | 9.15 | 9.33 | 10.0 | 9.25 | 7.67 | 9.5 | 18.9s | 12 | Accuracy |
| 4 | claude-opus-4-6 | 9.08 | 9.12 | 9.5 | 9.38 | 7.62 | 9.75 | 46.4s | 8 | Coherence |
| 5 | deepseek-r1 | 9.07 | 9.08 | 9.67 | 9.08 | 7.83 | 9.67 | 24.3s | 12 | Creativity |
| 6 | gpt-4-1-mini | 8.92 | 9.2 | 9.9 | 9.1 | 6.9 | 9.5 | 9.5s | 10 | Accuracy |
| 7 | gpt-5-4 | 8.91 | 9.0 | 9.78 | 9.11 | 7.11 | 9.56 | 9.8s | 9 | Coherence |
| 8 | llama-4-scout | 8.66 | 8.79 | 9.64 | 8.79 | 6.79 | 9.29 | 13.0s | 14 | Relevance |
| 9 | deepseek-ai/DeepSeek-R1-0528 | 8.63 | 8.67 | 9.67 | 8.67 | 7.17 | 9.0 | 33.6s | 6 | Relevance |
| 10 | Qwen/Qwen2.5-72B-Instruct | 8.60 | 8.86 | 9.71 | 8.71 | 6.57 | 9.14 | 19.3s | 7 | Relevance |
| 11 | mistral-medium | 8.60 | 8.5 | 9.38 | 8.88 | 7.0 | 9.25 | 36.6s | 8 | Coherence |
| 12 | Qwen/Qwen3-235B-A22B-Instruct-2507 | 8.60 | 8.5 | 9.5 | 9.0 | 6.83 | 9.17 | 15.6s | 6 | Coherence |
| 13 | deepseek-ai/DeepSeek-V3 | 8.60 | 8.75 | 9.62 | 8.62 | 7.0 | 9.0 | 24.8s | 8 | Relevance |
| 14 | grok-4-1-fast-reasoning | 8.47 | 8.44 | 9.33 | 8.0 | 7.67 | 8.89 | 15.9s | 9 | Creativity |
| 15 | phi-4 | 8.42 | 8.56 | 9.0 | 8.67 | 6.67 | 9.22 | 30.0s | 9 | Language Quality |
| 16 | openai/gpt-4o-mini | 8.42 | 8.5 | 9.33 | 8.67 | 6.25 | 9.33 | 8.7s | 12 | Language Quality |
| 17 | zai-org/GLM-4.6 | 8.17 | 7.83 | 8.83 | 8.33 | 6.83 | 9.0 | 20.5s | 6 | Creativity |
| 18 | meta-llama/Llama-4-Scout-17B-16E-Instruct | 8.16 | 8.2 | 9.0 | 8.4 | 6.0 | 9.2 | 14.2s | 5 | Language Quality |
| 19 | gpt-oss-120b | 8.16 | 8.11 | 9.0 | 7.89 | 6.78 | 9.0 | 8.4s ⚡ fastest | 9 | Creativity |
| 20 | openai/gpt-oss-120b | 8.04 | 8.2 | 8.6 | 8.4 | 6.0 | 9.0 | 27.6s | 5 | Language Quality |
Image Models
Scored 0–10 on accuracy, relevance, coherence, creativity, and image quality, plus the reviewer's overall rating. Image models rank with 3+ reviewed generations — image volume runs lower than text, and the sample size is shown per row.
| Model | Overall | Accuracy | Relevance | Coherence | Creativity | Image Quality | Avg Speed | Reviews | Strongest At | |
|---|---|---|---|---|---|---|---|---|---|---|
| 🥇 | gpt-image-1-mini | 9.05 | 9.25 | 9.75 | 9.0 | 8.0 | 9.25 | 32.8s | 4 | Accuracy |
| 🥈 | mai-image-2 | 8.88 | 9.0 | 9.6 | 9.0 | 7.6 | 9.2 | 26.9s | 5 | Accuracy |
| 🥉 | imagen3 | 8.63 | 8.33 | 8.83 | 9.0 | 7.5 | 9.5 | 10.3s | 6 | Image Quality |
| 4 | flux-dev | 8.40 | 7.4 | 8.8 | 8.8 | 7.4 | 9.6 | 10.1s | 5 | Image Quality |
Reliability
Quality only matters if the answer arrives. Success rate of generation jobs by AI platform on our infrastructure.
| Platform | Jobs | Success Rate |
|---|---|---|
| Azure AI | 6 | 100.0% |
| Black Forest Labs | 6 | 100.0% |
| 9 | 100.0% | |
| OpenAI | 10 | 80.0% |
| Microsoft | 8 | 62.5% |
How the Leaderboard Works
Real generations
Users run real prompts through real models — no synthetic test sets, no cherry-picking.
Independent review
Every output is scored 0–10 by an independent AI evaluator on five quality dimensions plus an overall rating.
Aggregated per model
Scores, latency, and success rates are averaged per model. Text models need 5+ reviews to rank, image models 3+.
Always current
Rankings shift with every new generation — this is live real-world output, not a one-time benchmark run.
Strongest At compares each model against the field average, so it highlights genuine standout ability rather than dimensions every model aces. For deeper matchup analysis, read our AI model comparisons and guides.
Last updated August 15, 2026.
Which model wins for your prompt?
Type it once — see every model's answer side by side, scored.
Compare Text Models Compare Image Models
