Size
Large
Tasks
97 scored · 3 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
WikiBench 1.0
100 real tasks for models running locally. Each task scores 0–10. The overall score is a weighted mean, and partial credit counts. A separate model judges the answers: Ollama qwen2.5:7b.
A separate judge, a local Qwen 2.5 7B, scored the rubric tasks. No model judged its own work.
Live rows from the public table. Cached for about a minute.
Size
Large
Tasks
97 scored · 3 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| GPT-6 Luna (cloud)CloudLarge | 4.98 | Large | 97 scored · 3 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Claude Haiku 5.5 (cloud)CloudLarge | 4.75 | Large | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) |
Size
Tiny
3B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Tiny
3.8B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Tiny
3B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Tiny
1.2B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Tiny
2.6B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Tiny
2.7B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (OpenRouter)
Size
Tiny
3B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| Granite 4.2 3B Q4 (local)Q4Local on SparkTiny | 4.30 | Tiny3B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Phi-4-mini Q4 (local)Q4Local on SparkTiny | 4.16 | Tiny3.8B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| SmolLM3 3B Q4 (local)Q4Local on SparkTiny | 4.14 | Tiny3B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| LFM2.5-1.2B Instruct Q4 (local)Q4Local on SparkTiny | 4.05 | Tiny1.2B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| LFM2.5-2.6B Q4 (local)Q4Local on SparkTiny | 3.81 | Tiny2.6B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Liquid LFM 2.5 2.6B (OpenRouter free)CloudTiny | 3.80 | Tiny2.7B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (OpenRouter) | ||
| Llama 3.2 3B Instruct Q4 (local)Q4Local on SparkTiny | 3.71 | Tiny3B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 |
Size
Small
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Small
8B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| Gemma 4 E4B Q4 (local)Q4Local on SparkSmall | 4.83 | Small | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| LFM2.5-8B-A1B Q4 (local)Q4Local on SparkSmall | 3.92 | Small8B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 |
Size
Medium
30B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Medium
24B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Medium
39.8B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (OpenRouter)
Size
Medium
33B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal, inferred)
Size
Medium
30B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (OpenRouter)
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| Nemotron 3.5 Lightning (local, TensorFold)NVFP4Local on SparkMedium | 4.97 | Medium30B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| LFM2.5-24B-A2B Q4 (local)Q4Local on SparkMedium | 4.61 | Medium24B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Flint v1.0 by Recursive SI (OpenRouter free)CloudMedium | 4.52 | Medium39.8B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (OpenRouter) | ||
| Laguna XS 2.1 (free)CloudMedium | 4.29 | Medium33B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal, inferred) | ||
| North Mini Code (OpenRouter free)CloudMedium | 4.26 | Medium30B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (OpenRouter) |
Size
Spark-sized
552B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Spark-sized
Tasks
97 scored · 3 unsupported (audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Spark-sized
Tasks
97 scored · 3 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Spark-sized
Tasks
97 scored · 3 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Spark-sized
122B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
ASUS Ascent GX10
Size
Spark-sized
118B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal, inferred)
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash EXL3 (local, thinking off, vcruz305 DERISKED pack)EXL3 1.62bpwLocal on SparkSpark-sized | 5.35 | Spark-sized552B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Qwen3.8 Flash Next (vcruz SAGE EXL3 4.77bpw, no thinking)EXL3 4.77bpwLocal on SparkSpark-sized | 5.21 | Spark-sized | 97 scored · 3 unsupported (audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Qwen 3.8 Flash Next EXL3 (local, ViC305)EXL3Local on SparkSpark-sized | 5.17 | Spark-sized | 97 scored · 3 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Qwen 3.8 Flash Next FP4 (local)FP4Local on SparkSpark-sized | 4.91 | Spark-sized | 97 scored · 3 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Ornith 1.5 122B Q4 (local)Q4Local on SparkSpark-sized | 4.84 | Spark-sized122B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | ASUS Ascent GX10 | ||
| Laguna S 2.1 (free)CloudSpark-sized | 4.66 | Spark-sized118B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal, inferred) |
Size
Large
552B
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
94 scored · 6 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
97 scored · 3 unsupported
Run date
Judge
qwen2.5:7b (local, ASUS Ascent GX10)
Hardware
Hosted (Nous Portal)
Size
Large
501B
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
93 scored · 7 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
280B
Tasks
95 scored · 5 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (OpenRouter)
Size
Large
428B
Tasks
94 scored · 6 unsupported (vision/audio)
Run date
Judge
Ollama qwen2.5:7b
Hardware
Ollama cloud
Size
Large
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
Size
Large
Tasks
100 scored · 0 unsupported
Run date
Judge
Ollama qwen2.5:7b
Hardware
Hosted (Nous Portal)
| Model | Overall | Size | Tasks | Run date | Judge | Hardware | Breakdown |
|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017 (cloud)CloudLarge | 5.00 | Large552B | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Longcat 2.5 Preview (Nous Portal free)CloudLarge | 4.98 | Large | 94 scored · 6 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Kimi K3.1 Pro (cloud, thinking off)CloudLarge | 4.85 | Large | 97 scored · 3 unsupported | qwen2.5:7b (local, ASUS Ascent GX10) | Hosted (Nous Portal) | ||
| Beam 1.0 (Nous Portal free)CloudLarge | 4.72 | Large501B | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Ling 3.1 Flash (cloud)CloudLarge | 4.31 | Large | 93 scored · 7 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Dots3-Note Preview (OpenRouter free)CloudLarge | 4.30 | Large280B | 95 scored · 5 unsupported (vision/audio) | Ollama qwen2.5:7b | Hosted (OpenRouter) | ||
| MiniMax M3.1 Flash (cloud)CloudLarge | 4.26 | Large428B | 94 scored · 6 unsupported (vision/audio) | Ollama qwen2.5:7b | Ollama cloud | ||
| MiMo V2.6 Flash (cloud)CloudLarge | 3.91 | Large | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| GLM 5.3 Flash (cloud)CloudLarge | 3.43 | Large | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Mercury 3 (cloud)CloudLarge | 2.99 | Large | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) | ||
| Tencent Hy3 (cloud)CloudLarge | 2.94 | Large | 100 scored · 0 unsupported | Ollama qwen2.5:7b | Hosted (Nous Portal) |
WikiBench 1.0 is the first live version of this benchmark. It has 100 tasks in 12 weighted categories: agentic, tool use, computation, coding, visualization, security, research/writing, hallucination, humanized, computer use, vision, transcription.
Size tiers go by total parameters: Tiny (under 4B), Small (4B to under 15B), Medium (15B to under 70B), Spark-sized (70B or more, with a quant that fits the DGX Spark's 128GB of unified memory, about 110GB of weights or less), and Large (anything bigger, plus cloud-only models with no public size). Parameter counts are shown only when they are public or the run reported them.
Unsupported tasks are counted and left out of the mean. They are not scored as zero. Models are ranked by their mean over the tasks they ran. Equal scores keep the order of the reviewed results list. If a run has not reported category scores, the breakdown says pending. This page does not fill those in.
A separate judge, a local Qwen 2.5 7B, scored the rubric tasks. No model judged its own work. Frontier Models are hosted models from frontier labs. They are listed in their own group, outside the size tiers. Runs that scored no tasks are left off the board instead of being ranked as zero.
The Spark runner posts each finished result file to a token-protected endpoint with scripts/wikibench_post_result.py. New results land as drafts in wikibench_results and appear here only after they are reviewed and published. The public key can only read. Site research notes live on the methodology page.