The AI Models Guide — What to Use for Every Task
A living guide to the AI model landscape: 50 tracked models grouped by task, weekly movement and the skills employers hire for — refreshed automatically.
11 min read
A living guide to the AI model landscape: 50 tracked models grouped by task, weekly movement and the skills employers hire for — refreshed automatically.
Quick Answer
Task-by-task guide to AI models, refreshed weekly from live data: 50 tracked models across LLMs, embeddings, vision and speech plus demand from 2,068 AI job lis
Search Snapshot
50 models tracked · snapshot 2026-07-05 · 2,068 active AI job listings analysed
The right model depends entirely on your task — a text generator built for conversation will perform poorly at semantic search, and an embedding model cannot caption an image. This guide tracks 50 models across 8 families, from text generation and embeddings to speech, computer vision and time series forecasting. Download counts are pulled from Hugging Face weekly, and job demand figures come from 2068 active AI listings so you can match model knowledge to hiring signals. Use the family sections to find the right tool first, then compare models within it.
Work through these in order — most wrong choices come from starting at step 4.
| Model family | Use it for | Current leader |
|---|---|---|
| Text generation and LLMs | Chat assistants, coding copilots, agents, summarization, drafting and any task where the output is free-form text. | Qwen3-0.6B |
| Vision-language and multimodal | Describing images, reading documents and screenshots, answering questions about visual content and combining text with other media. | gemma-4-26B-A4B-it |
| Embeddings, search and ranking | Semantic search, RAG retrieval, deduplication, clustering and recommendation. Rerankers refine an initial result list for precision. | all-MiniLM-L6-v2 |
| Language understanding and classification | Sentiment analysis, topic and intent classification, named-entity recognition, translation and other structured judgements about text. | bert-base-uncased |
| Speech and audio | Transcription, voice interfaces, dubbing, meeting notes and audio event detection. | Kokoro-82M |
| Computer vision | Classifying images, detecting and segmenting objects, visual quality control and content moderation. | mobilenetv3_small_100.lamb_in1k |
| Time series forecasting | Demand forecasting, anomaly detection and other predictions over numeric sequences and tables. | chronos-2 |
| Other specialist models | Niche tasks — protein folding, robotics, graph learning and research architectures that have not settled into a mainstream category. | electra-base-discriminator |
Use it for: Chat assistants, coding copilots, agents, summarization, drafting and any task where the output is free-form text.
Text generation and LLMs are dominated by Qwen this cycle, with Qwen3-0.6B pulling 28.0M downloads and Qwen3-8B adding another 16.0M — together outpacing Meta's facebook/opt-125m, which still holds third place at 13.8M downloads. The gap between Qwen's smallest and largest charting models suggests strong demand at both the edge-deployment and mid-range ends of the market.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| Qwen3-0.6B | Qwen | text-generation | 28.0M | 1.4k | +255.1k |
| Qwen3-8B | Qwen | text-generation | 16.0M | 1.2k | +2.5M |
| opt-125m | text-generation | 13.8M | 271 | +1.1M | |
| gpt2 | openai-community | text-generation | 13.2M | 3.3k | +238.8k |
| Qwen2.5-7B-Instruct | Qwen | text-generation | 12.8M | 1.4k | +119.1k |
| Qwen2.5-1.5B-Instruct | Qwen | text-generation | 11.2M | 759 | — |
The largest and fastest-moving family. Frontier models (Claude, GPT, Gemini) are API-only; the open-weight models listed here are what you run yourself.
Use it for: Describing images, reading documents and screenshots, answering questions about visual content and combining text with other media.
Vision-language and multimodal models are led entirely by Google's Gemma 4 family this period, with the 26B MoE variant gemma-4-26B-A4B-it reaching 13.6M downloads and gemma-4-31B-it close behind at 11.2M. Qwen2.5-VL-7B-Instruct takes third at 9.8M downloads, signalling that sub-10B instruction-tuned multimodal models retain significant traction among practitioners.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| gemma-4-26B-A4B-it | image-text-to-text | 13.6M | 1.2k | +395.7k | |
| gemma-4-31B-it | image-text-to-text | 11.2M | 3.1k | +109.8k | |
| Qwen2.5-VL-7B-Instruct | Qwen | image-text-to-text | 9.8M | 1.6k | +229.6k |
| Qwen3.5-9B | Qwen | image-text-to-text | 8.8M | 1.7k | — |
Use these when the input is not just text — document extraction, chart reading and UI understanding all live here.
Use it for: Semantic search, RAG retrieval, deduplication, clustering and recommendation. Rerankers refine an initial result list for precision.
Embeddings, search and ranking show the most lopsided download numbers across all categories, with sentence-transformers/all-MiniLM-L6-v2 recording 244.7M downloads — more than three times the combined total of the next two leaders. cross-encoder/ms-marco-MiniLM-L6-v2 (80.1M) and BAAI/bge-small-en-v1.5 (61.7M) confirm that small, fast retrieval models remain the backbone of production search pipelines.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| all-MiniLM-L6-v2 | sentence-transformers | sentence-similarity | 244.7M | 5.0k | — |
| ms-marco-MiniLM-L6-v2 | cross-encoder | text-ranking | 80.1M | 272 | — |
| bge-small-en-v1.5 | BAAI | feature-extraction | 61.7M | 503 | — |
| paraphrase-multilingual-MiniLM-L12-v2 | sentence-transformers | sentence-similarity | 47.8M | 1.3k | — |
| all-mpnet-base-v2 | sentence-transformers | sentence-similarity | 33.1M | 1.3k | — |
| bge-m3 | BAAI | sentence-similarity | 32.2M | 3.2k | +852.5k |
The workhorse family behind almost every RAG system. Small, cheap to run and rarely the bottleneck — pick one with strong retrieval benchmarks and move on.
Use it for: Sentiment analysis, topic and intent classification, named-entity recognition, translation and other structured judgements about text.
Language understanding and classification is still anchored by google-bert/bert-base-uncased, which logged 62.7M downloads despite being one of the oldest models in this guide. FacebookAI/xlm-roberta-base (20.4M) and google-t5/t5-small (19.1M) round out the top three, showing that multilingual classification and lightweight sequence-to-sequence tasks continue to drive steady hiring demand for NLP engineers.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| bert-base-uncased | google-bert | fill-mask | 62.7M | 2.7k | +2.5M |
| xlm-roberta-base | FacebookAI | fill-mask | 20.4M | 859 | — |
| t5-small | google-t5 | translation | 19.1M | 563 | +5.4M |
| bge-reranker-v2-m3 | BAAI | text-classification | 16.2M | 1.1k | — |
| roberta-large | FacebookAI | fill-mask | 12.8M | 303 | +697.7k |
| roberta-base | FacebookAI | fill-mask | 10.6M | 618 | — |
BERT-family encoders remain heavily used in production because they are fast, cheap and deterministic to fine-tune for a fixed label set.
Use it for: Transcription, voice interfaces, dubbing, meeting notes and audio event detection.
Speech and audio generation is led by hexgrad/Kokoro-82M at 13.9M downloads, a notably compact TTS model that edges out laion/clap-htsat-fused (12.8M downloads), which serves audio-text contrastive tasks rather than synthesis. coqui/XTTS-v2 sits third at 9.3M downloads, reflecting continued interest in multilingual voice cloning alongside audio understanding.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| Kokoro-82M | hexgrad | text-to-speech | 13.9M | 6.4k | — |
| clap-htsat-fused | laion | audio-classification | 12.8M | 108 | — |
| XTTS-v2 | coqui | text-to-speech | 9.3M | 3.6k | — |
| speaker-diarization-3.1 | pyannote | automatic-speech-recognition | 8.3M | 2.6k | — |
Speech-to-text (ASR) and text-to-speech are separate model families — most voice products chain one of each around an LLM.
Use it for: Classifying images, detecting and segmenting objects, visual quality control and content moderation.
Computer vision download volume is split between lightweight classification and CLIP-based retrieval, with timm/mobilenetv3_small_100.lamb_in1k leading at 26.0M downloads. openai/clip-vit-base-patch32 (22.2M) and openai/clip-vit-large-patch14 (12.5M) together show that zero-shot visual search remains one of the most actively deployed vision capabilities in production systems.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| mobilenetv3_small_100.lamb_in1k | timm | image-classification | 26.0M | 83 | +4.8M |
| clip-vit-base-patch32 | openai | zero-shot-image-classification | 22.2M | 969 | — |
| clip-vit-large-patch14 | openai | zero-shot-image-classification | 12.5M | 2.0k | +663.2k |
| nsfw_image_detection | Falconsai | image-classification | 9.1M | 1.1k | — |
Zero-shot vision models classify against labels you supply at runtime — no retraining needed when categories change.
Use it for: Demand forecasting, anomaly detection and other predictions over numeric sequences and tables.
Time series forecasting is consolidating around Amazon and AutoGluon's Chronos ecosystem, with amazon/chronos-2 recording 15.2M downloads and autogluon/chronos-bolt-small reaching 13.6M. autogluon/chronos-2 adds a further 7.9M downloads, making Chronos-family models the clear reference point for practitioners entering the forecasting job market.
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| chronos-2 | amazon | time-series-forecasting | 15.2M | 345 | — |
| chronos-bolt-small | autogluon | time-series-forecasting | 13.6M | 45 | — |
| chronos-2 | autogluon | time-series-forecasting | 7.9M | 33 | New |
Foundation models for time series are a recent arrival — they forecast series they were never trained on, the way LLMs handle unseen text.
Use it for: Niche tasks — protein folding, robotics, graph learning and research architectures that have not settled into a mainstream category.
Other specialist models are led by google/electra-base-discriminator at 40.0M downloads, a figure that rivals mainstream classification models and reflects its widespread use in efficient text discrimination tasks. Bingsu/adetailer (11.9M) and colbert-ir/colbertv2.0 (8.9M) cover image inpainting detection and late-interaction retrieval respectively, illustrating the
| Model | Maker | Task | Downloads | Likes | This week |
|---|---|---|---|---|---|
| electra-base-discriminator | — | 40.0M | 132 | — | |
| adetailer | Bingsu | — | 11.9M | 736 | — |
| colbertv2.0 | colbert-ir | — | 8.9M | 363 | — |
This week's snapshot, dated 2026-07-05, adds three models to the tracked set: sentence-transformers/paraphrase-multilingual-mpnet-base-v2, deepseek-ai/DeepSeek-R1 and autogluon/chronos-2. The fastest-growing model by raw download volume is google-t5/t5-small, which added 5.4M downloads in a single week, followed by timm/mobilenetv3_small_100.lamb_in1k at +4.8M and intfloat/multilingual-e5-large at +3.6M.
| New in the tracked set | Task | Downloads |
|---|---|---|
| sentence-transformers/paraphrase-multilingual-mpnet-base-v2 | sentence-similarity | 8.5M |
| deepseek-ai/DeepSeek-R1 | text-generation | 8.2M |
| autogluon/chronos-2 | time-series-forecasting | 7.9M |
| Fastest download growth | Task | Added this week |
|---|---|---|
| google-t5/t5-small | translation | +5.4M |
| timm/mobilenetv3_small_100.lamb_in1k | image-classification | +4.8M |
| intfloat/multilingual-e5-large | feature-extraction | +3.6M |
| Qwen/Qwen3-8B | text-generation | +2.5M |
| google-bert/bert-base-uncased | fill-mask | +2.5M |
| facebook/opt-125m | text-generation | +1.1M |
| BAAI/bge-m3 | sentence-similarity | +852.5k |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | text-generation | +825.5k |
Machine Learning skills appear in 642 job listings — the highest count of any tracked skill — while LLMs and GenAI follow at 440 listings, reflecting strong employer appetite for both foundational and generative model knowledge. RAG appears in 94 listings, directly connecting the embeddings, search and ranking family (13 models tracked, the largest family in this guide) to real hiring demand. Python underpins nearly all of this work at 258 listings and PyTorch sits at 118, making framework fluency as marketable as model-specific expertise.
| Skill | Active listings |
|---|---|
| Machine Learning | 642 |
| LLMs / GenAI | 440 |
| Python | 258 |
| PyTorch | 118 |
| Stakeholder Mgmt | 111 |
| RAG | 94 |
| TensorFlow | 80 |
| GCP | 66 |
| SQL | 59 |
| Kubernetes | 58 |
| NLP | 57 |
| Statistical Analysis | 55 |
Explore the full demand data on the AI and ML job market page or track movement over time on skill trends.
Model popularity comes from a weekly snapshot of Hugging Face download and like counts (50 top models, latest snapshot 2026-07-05). "This week" columns compare the latest snapshot with one taken roughly a week earlier, so they show download velocity rather than all-time totals. Employer demand comes from daily analysis of active AI job listings (2,068 at last count). Frontier API-only models do not appear in download tables because they publish no open weights — that is a limitation of the source, not a quality judgement.
For RAG and semantic search, the embeddings family is the right starting point — sentence-transformers/all-MiniLM-L6-v2 leads the entire tracked set with 244.7M downloads, signalling broad production adoption. intfloat/multilingual-e5-large is also gaining ground fast, adding 3.6M downloads this week alone, making it a strong candidate when multilingual retrieval matters.
The most downloaded AI model in this snapshot is sentence-transformers/all-MiniLM-L6-v2 with 244.7M total downloads on Hugging Face. It sits in the embeddings, search and ranking family and is widely used for semantic similarity, clustering and retrieval tasks.
Machine Learning is the most requested skill across the 2068 active job listings tracked, appearing in 642 postings, with LLMs and GenAI close behind at 440. Python appears in 258 listings and PyTorch in 118, while RAG — a direct application of embedding models — shows up in 94 listings, rewarding candidates who can connect model knowledge to retrieval system design.
Updated July 10, 2026. This guide regenerates automatically from fresh data every week.
Actionable guides, market updates and shipping notes — once a week.