exploring AI models | github.com/Moh4696

dm is open
this 17-year-old is the founder of Cal AI, an app making over $1M a month. Cal AI has reached 15M+ downloads, and one of its biggest growth channels is AI UGC ads. the article below was written by Cal AI’s co-founder, breaking down step-by-step exactly how they create AI UGC ads. bookmark this one.
Paid partnership (ad)
2
1
12
1,673
🚨 Opus 5.5 just built this in one hour. a full Arabian luxury house 3D experience built with Three.js. the crazy part? i didn’t give it a single reference image. just the prompt. and it still came up with the entire visual direction from scratch. view it here: moh4696.github.io/qasr-al-no…
🚨Claude Opus 5.5 VS GPT 6.1 Sol. i gave both models the same simple prompt: create an animated website from a single image. Opus took around 1hr 20mins, while GPT-6.1 finished in just 15mins. both came up with really nice concepts, but Opus clearly spent more time on the details and delivered the better overall quality. GPT-6.1 wins on speed. Opus wins on execution. you can check both websites in the comments and decide for yourself.
10
2
39
2,764
m0h retweeted
🚨Claude Opus 5.5 VS GPT 6.1 Sol. i gave both models the same simple prompt: create an animated website from a single image. Opus took around 1hr 20mins, while GPT-6.1 finished in just 15mins. both came up with really nice concepts, but Opus clearly spent more time on the details and delivered the better overall quality. GPT-6.1 wins on speed. Opus wins on execution. you can check both websites in the comments and decide for yourself.
gave the same prompt to these four models and here's their result. opus 5.5 overall was the best performer, the other three models including GPT 6.1. sol messed it up pretty bad. anthropic really hacked something with this 5.5 series, fable 5.5 will be madder!!!
27
10
144
18,865
google winning is pure data gravity. workspace gives Gemini an unmatched dataset of context.
Which personal agent is the best life assistant? Today we're releasing the micro1 PersonalAgentBench to help answer that, and we're paying real personal agent users to contribute to the benchmark. Personal agents can now book your flights, answer emails, and move your meetings around. That’s great, until one messages the wrong person or makes a purchase without confirming. Capability is not the same as trust. So we put four personal agents to the test: Instinct, Muse, Grok Bot, and Gemini Spark. We found that even when agents found the right information, they still often miss what matters, overshare, or make things up. To find out for yourself how your agent compares, you can join our benchmark. We’re paying the first 100 users to run our prompts and share their results.
1
1
21
2,646
this repo is trending at #2 on GitHub and has already reached 278k+ stars. it contains 38 real developer skills designed to supercharge your coding agents. the skills are based on decades of engineering experience and were built by Matt Pocock. and the best part? setup takes about 30 seconds. repo link is in the comments.
this open-source agent harness is the #1 trending tool on GitHub right now. that’s because it beat OpenClaw and Hermes on efficiency. it’s built to be more lightweight, uses less RAM & CPU, and delivers results much faster. 🔗: github.com/tinyhumansai/open…
1
24
2,492
you can now use AI with more privacy and control. antseed just raised $2.4M to build an open AI marketplace where you can access models, choose providers and pay directly for what you use. no sign-up, competitive pricing, and private inference through TEE providers.
What BitTorrent did for files, we want to do for AI inference. Antseed lets anyone buy and sell access to AI models, on their own terms. The Antseed Foundation has raised $2.4M in a token round led by Spark Capital to help make that happen.
Paid partnership (ad)
3
19
2,281
🚨Claude Opus 5.5 VS GPT 6.1 Sol. i gave both models the same simple prompt: create an animated website from a single image. Opus took around 1hr 20mins, while GPT-6.1 finished in just 15mins. both came up with really nice concepts, but Opus clearly spent more time on the details and delivered the better overall quality. GPT-6.1 wins on speed. Opus wins on execution. you can check both websites in the comments and decide for yourself.
gave the same prompt to these four models and here's their result. opus 5.5 overall was the best performer, the other three models including GPT 6.1. sol messed it up pretty bad. anthropic really hacked something with this 5.5 series, fable 5.5 will be madder!!!
27
10
144
18,865
Australia finally joins the AI race, and it’s open-source. 🇦🇺 it’s a 26B parameter model with a reasoning score of 44.09. it’s already available on Hugging Face. link in the comments.
Today we are open-sourcing Matilda Jev, Australia's first decision model, conditioned for Australian deployment and post-trained with our own specialised recipe. Decision models do not generate text. They answer a typed question and respond with a calibrated probability for each option in a single forward pass, reducing hallucination and making intelligent decisions faster and more efficient. This release proves that highly capable AI models can be built, post-trained, evaluated, and served entirely within Australia. Post-trained end-to-end on our local infrastructure, the model achieved a median decision latency of 56.8ms on a single @AMD MI355X across 150,317 evaluation requests.
Paid partnership (ad)
2
18
2,248
this open-source agent harness is the #1 trending tool on GitHub right now. that’s because it beat OpenClaw and Hermes on efficiency. it’s built to be more lightweight, uses less RAM & CPU, and delivers results much faster. 🔗: github.com/tinyhumansai/open…
FOUND A @NousResearch and @openclaw KILLER. FASTER AND CHEAPER THAN BOTH Saw this and had to test it myself: → Summarize my emails → The exact same prompt → Same inbox, same task Ran them side by side and the speed gap is wild. Plus OpenHuman shows you: → Your context window live → What each session costs After testing both on real work, @tinyhumansai - openhuman is my agent now.
26
195
1,504
125,230
gave the same prompt to these four models and here's their result. opus 5.5 overall was the best performer, the other three models including GPT 6.1. sol messed it up pretty bad. anthropic really hacked something with this 5.5 series, fable 5.5 will be madder!!!
17
2
46
20,008
this is the full and ref image that I used: You are a senior real-time 3D artist and Three.js engineer. Build a detailed, realistic, interactive mid-engine supercar in the existing project using Three.js and WebGL. Use the attached image as a visual reference. It shows one original, modern Italian-inspired supercar from front three-quarter, side, and rear three-quarter angles. Keep the design consistent across all views. Do not reproduce the image as a flat plane, billboard, or texture. Build a real 3D car that looks convincing from every angle. Work only inside this project folder. I do not need a mobile view or mobile-specific layout. Focus on the reference image and this prompt. Required workflow: Complete the phases below in order. At the end of every phase, run the project using its existing workflow, inspect the actual rendered result, fix problems, and repeat the checks until the phase passes. Keep the project runnable between phases. Do not mark a phase complete based on code inspection alone when its appearance or behavior needs to be checked in a rendered scene. Make routine implementation choices yourself and continue through all phases without waiting for my review. Phase 0 — Inspect the project Inspect the project structure, Three.js version, rendering setup, conventions, local assets, and run and preview workflow. Decide where the reusable car component belongs and how it should integrate. Reuse suitable local assets and existing patterns. Do not add external dependencies or fetch assets unless the project already has an established workflow for doing so. Check: Confirm that the implementation fits the project architecture and that you know how to launch and inspect the result. Phase 1 — Build the silhouette Create a low, wide, wedge-shaped mid-engine supercar with a short nose, swept windshield, compact two-seat cabin, broad rear shoulders, deep side intakes, and short overhangs. Use approximate real-world scale: 4.5 m long, 2.0 m wide, and 1.2 m tall. Use Y as up, X as left/right, and +Z as forward. Place the origin at the car’s ground-level center, or document a different origin if required by the project. Create actual shaped 3D geometry. Do not leave the result as a crude collection of boxes or floating primitives. Check: Inspect front three-quarter, side, and rear three-quarter views. Correct the silhouette, proportions, wheelbase, and wheel placement. Phase 2 — Build the exterior Create separate body panels with readable gaps around the doors, front storage lid, and rear engine hatch. Add performance tires with visible tread, detailed wheels, brake discs and calipers, headlights, tail lights, indicators, mirrors, window glass, wipers, front splitter, side skirts, rear diffuser, exhaust outlets, grilles, and visible underbody details. Make the side and rear air intakes look functional and integrated into the body. Check: Inspect the same three views. Correct visible intersections, disconnected parts, and missing surfaces. Phase 3 — Refine materials Match the reference image’s metallic burnt-orange paint, dark wheels, restrained carbon-fiber accents, clear glazing, and realistic lights. Use plausible roughness and reflections for paint, glass, rubber, metal, leather, and carbon fiber. Keep panel edges and details readable under ordinary scene lighting. Avoid excessive gloss, oversaturated colors, and unstable shaders. Check: Inspect the car under the project’s normal lighting from front, side, and rear angles. Adjust materials until its form remains clear. Phase 4 — Build the cockpit and engine bay Create a complete, believable two-seat interior visible through the windows and open doors. Include a shaped dashboard, readable instrument display, steering wheel and column, central controls, pedals, seats with bolsters and seams, seat belts, inner door panels, floor, and center tunnel. Build a convincing engine assembly beneath the rear hatch. Include engine covers, intake parts, hoses, wiring, cooling components, exhaust routing, and supporting structure. The components need not simulate a particular production engine, but their scale, arrangement, and fit must look credible. Check: Inspect the interior through closed windows and open doors, then inspect the open engine bay. Correct clipping, misplaced parts, and glass that hides the interior. Phase 5 — Animate the doors and hatches Make both doors open and close smoothly using mechanically plausible pivots. Scissor-style doors are suitable if the hinge motion and clearances are credible. Make the rear engine hatch operable, and make the front storage lid operable if the design includes one. Keep mirrors, glass, handles, trim, and inner panels attached to their moving parts. Expose a clear API for toggling and explicitly opening or closing each part. Add interaction targets or raycasting hooks for a host application. Keep optional demo controls in the demo page rather than imposing UI on the reusable car asset. Check: Open and close every moving part repeatedly. Verify pivots, easing, clearances, and attachments. Phase 6 — Add engine audio Provide engine start, idle, rev, and stop behavior if the runtime supports browser audio. Prefer suitable local audio assets. Otherwise, create layered Web Audio synthesis with harmonic content, filtering, and subtle noise. Revving should smoothly change pitch, intensity, and texture. Initialize or resume audio only after a user gesture. Never autoplay. Handle muted or unavailable audio gracefully, and stop and dispose of audio nodes cleanly. Expose start, stop, and optional throttle or rev controls through the car’s API. Check: Test start, idle, rev, stop, and disposal through user actions. Confirm there is no autoplay, lingering audio, or console error. Phase 7 — Integrate and polish Keep the car asset self-contained and separate from unrelated scene setup. Do not bake in a floor, background, scenery, camera, or permanent environment lighting. Its materials should work with lighting supplied by a host scene. Document a concise interface for creating the car, attaching its root object to a scene, controlling doors and hatches, controlling engine audio, updating animation, and disposing of resources. Run the project and inspect front three-quarter, side, rear three-quarter, cockpit, and open-engine views. Refine proportions, geometry, materials, and details until the result closely follows the attached reference and remains convincing from every angle. Final report: List the files changed, explain how to instantiate and control the car, state which visual and audio checks you completed, and clearly mention any checks you could not perform. Do not stop at a plan or blockout.
6
601
this is way bigger than an AI email client. Hedwig is building one visual workspace that connects your tools, gives an AI agent the full context, and lets it actually take actions for you. email, calendar, tasks, meetings, workflows. all in one place. built by people from Stanford & Meta. they’re giving out free access right now: hedwig-ai.com/r/ars
we’re going to kill a one trillion dollar industry. we’re building Hedwig: one visual workspace that connects your tools and gives your agent the full picture, and going after the biggest players like google and meta. built by a team from stanford, meta, and polymarket. loved by 6,000+ across the world comment + RT for free access hedwigmail.com
Paid partnership (ad)
8
1
25
3,527
capcut pro costs $19.99/month. here’s an open-source alternative that’s completely FREE. 91.3K GitHub stars. MIT licensed. works on web, desktop & mobile. you can handle all the basics: > import & export high-quality video and audio > multi-track timeline editing > add masks and export without watermarks and they’re currently rebuilding the tool with even more coming, including AI agents, plugins, and automated workflows. the GitHub repo is linked below.
grok-bot just got open-sourced and it run locally completely FREE. 3.8k stars in 2 weeks. MIT licensed. it gives every AI agent its own computer. own browser, own files, own container. every action is approved against a policy before it runs, and written to an audit log after. here's how to set it up: 1️⃣prerequisites Docker + Bun 1.3+ 2️⃣clone and create your env git clone github.com/CopilotKit/OpenBo… cd OpenBot cp .env.example .env 3️⃣get intelligence credentials (free plan works) npx --yes copilotkit@latest login npx --yes copilotkit@latest project select npx --yes copilotkit@latest license --write drop the cpk-... key into .env as INTELLIGENCE_API_KEY 4️⃣ add your model key "OPENAI_API_KEY=sk-..." 5️⃣ run it bun install bash scripts/start.sh open localhost:3010. that's it. the underrated part: the repo ships a prompt.txt you paste into claude code or cursor and it walks the whole setup for you. every claim in it is verified against the actual repo. what it actually costs: the code is MIT, free forever. you pay your own model tokens. intelligence has a free tier and can be fully self-hosted if you want zero external dependency. works with any AG-UI agent. langgraph, mastra, crewai, pydantic ai. repo in in the CS.
7
2
22
2,764