Just want to ask?
There are a lot of numbers on this page. Type your question and we'll point you to the answer, or text it to PJ and a real person answers.
Matches your words to SideGuy's live local pages, in your browser. Nothing is sent unless you tap Text PJ.
15 AI models from established labs are trending this week on Hugging Face, the public library where AI labs post their models. Most people never need to download one: new models show up inside the assistants already on your phone. This page says what each one does, whether a regular computer can run it, and who it actually helps. Read Wednesday, October 21.
The smartest AI models right now
Ranked by the Artificial Analysis Intelligence Index, which rolls hard tests of reasoning, knowledge, coding and agent work into one score. Each model is shown at its best setting. Kimi K3 ranks #21 of 479.
| # | Model | Index | API price per 1M tokens | Speed (tokens a second) |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 Anthropic, released 2026-09-22 | 57.6 | $4 in / $20 out | 96 |
| 2 | Claude Sonnet 5.5 Anthropic, released 2026-09-28 | 56.0 | $2 in / $10 out | 139 |
| 3 | Muse Wonder Full Meta, released 2026-10-12 | 55.8 | not listed | n/a |
| 4 | Claude Fable 5.1 Anthropic, released 2026-09-01 | 53.4 | $10 in / $50 out | 72 |
| 5 | GPT-6 Astra OpenAI, released 2026-09-03 | 52.7 | $10 in / $50 out | 66 |
| 6 | Gemini 4 Argon Google, released 2026-09-30 | 52.6 | $2 in / $10 out | 80 |
| 7 | Muse Wonder Meta, released 2026-10-12 | 52.4 | $4 in / $20 out | 172 |
| 8 | GPT-6.1 Sol OpenAI, released 2026-09-29 | 51.8 | $2 in / $10 out | 54 |
| 9 | Claude Opus 5 Anthropic, released 2026-07-24 | 50.8 | $5 in / $25 out | n/a |
| 10 | Claude Fable 5 Anthropic, released 2026-06-09 | 49.6 | $10 in / $50 out | n/a |
| 11 | Muse Spark 1.3 Meta, released 2026-09-02 | 48.1 | $1.25 in / $4.25 out | 123 |
| 12 | GPT-6 Sol OpenAI, released 2026-09-22 | 47.6 | $2 in / $10 out | n/a |
Data: Artificial Analysis, read every morning. Higher is better. Prices are what each company charges developers; the apps most people use (ChatGPT, Claude, Gemini) have their own monthly plans.
Trending this week, in plain English
| Model and lab | What it does | Can my computer run it? | License | Downloads |
|---|---|---|---|---|
| DeepSeek-V4.1-Flash-1017 DeepSeek | reads photos, screenshots and documents, then answers about them | 763B: data-center GPUs, or a hosted API | open, business use OK (mit) | 42,122 |
| flux-3-image-klein-comfy Black Forest Labs | edits or restyles a picture | size not listed: see the model card | open weights, read its license (other) | 490,447 |
| flux-3-image-klein Black Forest Labs | edits or restyles a picture | 6B: a regular laptop can run it with free apps like ComfyUI | open weights, read its license (other) | 92,604 |
| embeddinggemma-2 | powers search behind the scenes | 1B: a regular laptop can run it | open, business use OK (apache-2.0) | 179,734 |
| LTX-2.5 Lightricks | turns a picture into a short video | size not listed: see the model card | open weights, read its license (other) | 2,057,942 |
| Qwen3.8-27B Alibaba (Qwen) | reads photos, screenshots and documents, then answers about them | 28B: a strong desktop with a big graphics card | open, business use OK (apache-2.0) | 7,729,432 |
| Qwen3.8-Flash-Next-FP4-Local NVIDIA | reads photos, screenshots and documents, then answers about them | 96B: data-center GPUs, or a hosted API | open weights, read its license (other) | 9,340 |
| LFM2.5-24B-A2B Liquid AI | writes, answers questions and codes | 24B: a strong desktop with a big graphics card | open weights, read its license (other) | 16,939 |
| Hy-Memory-30B-A3B Tencent | writes, answers questions and codes | 30B: a strong desktop with a big graphics card | open, business use OK (apache-2.0) | 2,682 |
| Unlimited-OCR Baidu | reads photos, screenshots and documents, then answers about them | 3B: a regular laptop can run it with free apps like Ollama or LM Studio | open, business use OK (mit) | 2,127,860 |
| Qwen-Image-2.1-Turbo Alibaba (Qwen) | makes pictures from a description | 7B: a regular laptop can run it with free apps like ComfyUI | open weights, read its license (other) | 29,826 |
| NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 NVIDIA | reads photos, screenshots and documents, then answers about them | 66B: data-center GPUs, or a hosted API | open weights, read its license (other) | 12,910 |
| Qwen3.8-Flash-Next Alibaba (Qwen) | reads photos, screenshots and documents, then answers about them | 180B: data-center GPUs, or a hosted API | open weights, read its license (other) | 2,633,567 |
| DeepSeek-V4.1-Flash DeepSeek | reads photos, screenshots and documents, then answers about them | 763B: data-center GPUs, or a hosted API | open, business use OK (mit) | 1,468,263 |
| NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 NVIDIA | writes, answers questions and codes | 4B: a regular laptop can run it with free apps like Ollama or LM Studio | open weights, read its license (other) | 3,062 |
From Hugging Face's public model data, trending first, filtered to established labs (we leave out hobby uploads and models built to dodge safety rules or AI detectors). Size is the total parameter count; models that wake only part of themselves per answer can run lighter than it suggests.
The big open release: Kimi K3
Moonshot AI calls Kimi K3 "an open-weight, native multimodal agentic model" for long coding jobs, knowledge work and reasoning: 2.8 trillion parameters with 104 billion active per answer, a 1 million token memory, and text, images and video in one model. On its API, kimi-k3 costs $3.00 per million input tokens ($0.30 cached) and $15.00 per million output. Who it helps: developers and companies that want a frontier model they can host themselves. Who it doesn't: anyone hoping to run it at home; it needs data-center GPUs. Model card · Kimi pricing · how it compares on price with the coding agents.
Read October 14, 2026 on Moonshot's own pages.
Which one should I use?
- The assistant you already have. ChatGPT, Claude, Gemini or the one built into your phone. New models arrive inside these apps, so for most people this is the whole answer.
- Free and private, on your own computer. Small open models (under about 9 billion parameters) run on a regular laptop with free apps like Ollama or LM Studio, and nothing leaves the machine.
- A business wiring AI into its work. Hosted APIs from the big labs, with rules about where your data goes. Check the license before you build on an open model.
- Hand it to PJ. He sets it up and shows you how to use it, one on one or with your team. Take your AI to the next level · Text PJ.
Need a human?
This page is built every morning from public data by SideGuy, run by PJ, a local in Solana Beach. If something looks off, or you want a hand with your plans, text PJ at 858-461-8054. A real person answers.
Local help, free to read: make your fix plan · senior benefits checkup · lower your SDG&E bill · Medicare Extra Help · every 2027 Medicare plan · is my medication recalled? · the next 90 days. Plans start as low as $25. Your parent lives here and you don't? The Parent Check-In Plan, $25 a month.
Questions people ask
Which AI model is the smartest right now?
On the Artificial Analysis Intelligence Index today: Claude Opus 5.5 (Anthropic, 57.6), then Claude Sonnet 5.5 (Anthropic, 56.0) and Muse Wonder Full (Meta, 55.8). The ranking changes as new models come out; this page reads it every morning.
What is Kimi K3?
Moonshot AI's open-weight multimodal model: 2.8 trillion total parameters with 104 billion active, a 1 million token context window, and text, images and video in one model. Its API model kimi-k3 costs $3.00 per million input tokens and $15.00 per million output tokens, per Moonshot.
Can I run a new AI model on my own computer?
Small ones, yes: models under about 9 billion parameters run on a regular laptop with free apps like Ollama or LM Studio. Mid-size ones need a desktop with a big graphics card, and the largest need data-center GPUs or a hosted API.
Which new AI model should I use?
For most people, the assistant already on their phone or computer: new models show up inside ChatGPT, Claude, Gemini and the rest. Downloading a model yourself makes sense for privacy or for building something.
Where does this list come from?
Hugging Face's public model data, trending first, filtered to established AI labs and rebuilt every morning.
Know someone who needs this?
Got a question about this? Ask PJ a question. A real person in Solana Beach texts you back.
Something changed here? Text PJ the fix: if it changes this page, you get $5. How it works.
Spotted something new around town? Text it in. What do you want to see here? Ask PJ for a page. No login, a real person reads it.
More free local tools: the North County toolkit.