Smoke dataset
A fixture dataset that exercises the ingestion, validation and freshness machinery.
4 tracked
Directory
Language and embedding models
This directory tracks 18 language, embedding and speech models across Hugging Face Hub download activity and OpenRouter context and pricing. Every figure carries the source that reported it and the moment it was observed. Download counts are a rolling 30-day window, not a lifetime total. Values that fail validation are withheld rather than shown.
Live
191 fields
- huggingface
- 89/89 current
- just now
- openrouter
- 48/48 current
- just now
Moved since last reading
- Gemma 2 9B Instruct
- +122,272
- 864,175
- Llama 3.1 8B Instruct
- +118,113
- 6,225,344
- BGE-M3
- −50,628
- 37,229,647
- Whisper large-v3
- +34,619
- 4,712,143
Stars from Hugging Face Hub, OpenRouter model catalogue. Movement is measured between two real observations; values that failed validation are withheld rather than shown.
- Models
- 18
- On Hugging Face
- 13/18
- On OpenRouter
- 12/18
- Last sync
- just nowsucceeded
Gemma 2 9B InstructDownloads, 30 days864,175+122,272Hugging Face Hub · just nowNomic Embed Text v1.5Downloads, 30 days14,308,511−28,958Hugging Face Hub · just nowQwen2.5 Coder 32B InstructDownloads, 30 days1,215,982−2,694Hugging Face Hub · just nowMixtral 8x7B InstructDownloads, 30 days217,408−183Hugging Face Hub · just nowDeepSeek V3Downloads, 30 days1,134,368−13,832Hugging Face Hub · just nowLlama 3.1 8B InstructDownloads, 30 days6,225,344+118,113Hugging Face Hub · just nowPhi-4Downloads, 30 days624,150−2,430Hugging Face Hub · just nowDeepSeek R1Downloads, 30 days815,126+20,929Hugging Face Hub · just nowMistral 7B Instruct v0.3Downloads, 30 days2,332,647−20,998Hugging Face Hub · just nowWhisper large-v3Downloads, 30 days4,712,143+34,619Hugging Face Hub · just nowLlama 3.3 70B InstructDownloads, 30 days917,910+4,408Hugging Face Hub · just nowQwen2.5 72B InstructDownloads, 30 days326,043−14,690Hugging Face Hub · just now
No hiring column here. The job-board classifier extracts framework names, so the join can only speak about frameworks — for these entities we have not measured demand, rather than measured it to be zero.
| Select to compare | Model | Downloads/30d | Likes | Context | In $/M | Modality | Licence | 7-day downloads | Checked |
|---|---|---|---|---|---|---|---|---|---|
| Llama 3.1 8B InstructMeta | 6,225,344 | 7,861 | 131,072 | 0.05 | text | llama-3.1 | 7-day downloads: 7 readings, 5.9M to 6.2M. | Currentjust now | |
| Mistral 7B Instruct v0.3Mistral AI | 2,332,647 | 3,531 | — | — | text | apache-2.0 | 7-day downloads: 7 readings, 2.4M to 2.3M. | Currentjust now | |
| DeepSeek V3DeepSeek | 1,134,368 | 4,236 | 163,840 | 0.32 | text | deepseek | 7-day downloads: 7 readings, 1.1M to 1.1M. | Currentjust now | |
| Llama 3.3 70B InstructMeta | 917,910 | 3,058 | 131,072 | 0.1 | text | llama-3.3 | 7-day downloads: 7 readings, 924.1K to 917.9K. | Currentjust now | |
| Gemma 2 9B InstructGoogle | 864,175 | 954 | — | — | text | gemma | 7-day downloads: 7 readings, 757.3K to 864.2K. | Currentjust now | |
| Phi-4Microsoft | 624,150 | 2,309 | 16,384 | 0.07 | text | mit | 7-day downloads: 7 readings, 693.3K to 624.1K. | Currentjust now | |
| Qwen2.5 72B InstructAlibaba | 326,043 | 994 | 32,768 | 0.36 | text | qwen | 7-day downloads: 7 readings, 395.7K to 326.0K. | Currentjust now | |
| Mixtral 8x7B InstructMistral AI | 217,408 | 4,755 | — | — | text | apache-2.0 | 7-day downloads: 7 readings, 217.1K to 217.4K. | Currentjust now | |
| Claude Opus 5Anthropic | — | — | 1,000,000 | 5 | text | proprietary | No readings | not checked | |
| Claude Sonnet 5Anthropic | — | — | 1,000,000 | 2 | text | proprietary | No readings | not checked | |
| Gemini 3.7 FlashGoogle | — | — | 1,048,576 | 0.75 | text | proprietary | No readings | not checked | |
| GPT-4oOpenAI | — | — | 128,000 | 2.5 | text | proprietary | No readings | not checked |
1 of 3 models selected — pick one more to compare
CompareClear
Sources: Hugging Face Hub, OpenRouter model catalogue. Every value carries the moment it was observed; figures that failed validation are withheld rather than shown. SLA: current within 24h, ageing to 96h, stale beyond 240h.
indexable · canonical /data/models?modality=text
Also tracked
Other datasets
Same engine, same provenance rules — every figure carries the source that reported it and when.
Agent engineering frameworks
Open-source frameworks for building, orchestrating and evaluating LLM agents, tracked across GitHub, npm and PyPI.
18 tracked
Local inference hardware
GPUs, accelerators and SoCs for running models locally, with VRAM, memory bandwidth, power draw and list price.
10 tracked