Alibaba Cloud's decision-model-preview is now available on Eden AI 🚀 It joins TypeSafe AI's Jev as the second System One model in our catalogue, and both speak the same Choice, Score and Noul primitives: you send a state and typed questions, you get back a typed answer with a probability. Nothing to parse. 👉 Why decision models are their own category: https://lnkd.in/enxXsCDx 👉 Try both and compare them from one API: https://lnkd.in/etHQzj4h
Eden AI
Développement de logiciels
Lyon, Auvergne-Rhône-Alpes 13 522 abonnés
AI made easy for Developers
À propos
Eden AI simplifies the use and deployment of AI technologies by providing a unique API connected to the best AI engines
- Site web
-
https://www.edenai.co
Lien externe pour Eden AI
- Secteur
- Développement de logiciels
- Taille de l’entreprise
- 11-50 employés
- Siège social
- Lyon, Auvergne-Rhône-Alpes
- Type
- Société indépendante
- Fondée en
- 2022
- Domaines
- Artificial Intelligence, API, Cloud, Software-as-a-Service, Computer Vision, Natural Language Processing, Machine Translation, Speech-to-Text, Machine Learning et Optical Character Recognition
Lieux
-
Principal
Obtenir l’itinéraire
69003 Lyon, Auvergne-Rhône-Alpes, FR
Employés chez Eden AI
Nouvelles
-
What if an AI model didn’t need to generate text at all? Jev, developed by TypeSafe AI, takes a different approach: it is built to make fast, structured decisions instead of generating prose. Through Eden AI’s new Decisions API, Jev can now be used for tasks such as classification, routing, scoring, gating, and deciding whether a more capable model needs to run. Instead of generating text and asking your application to interpret it, you define the decision you need and get back a structured result with probabilities and confidence. We explore the idea behind Jev, its Choice, Score and Noul primitives, and how decision-focused models can fit into modern AI systems. 🛠️ Try Decisions with Jev through Eden AI: https://lnkd.in/gAU5cWFk 🔗 Full article in the first comment.
-
-
Latency is the reason most agents skip grounding altogether. A real-time assistant cannot pause for a slow search, so it answers from memory and guesses instead. Sub-200ms changes what you can afford to ground. Linkup is already available on Eden AI for search, scraping and research, served from an EU region. Congratulations to the team. ⚡
Today we’re launching Flash, the most accurate web search API under 200 milliseconds. 190ms median latency. 295ms at the 90th percentile. The highest accuracy on SimpleQA in this latency range. Flash brings live web knowledge to agents that previously couldn’t afford slow latency: real-time assistants can now ground their answers in current information without disrupting the conversation. It also enables high-throughput reinforcement learning and small language model workflows that depend on fast, repeated access to external knowledge. This mode runs on our own index and search infrastructure, which is highly optimized for efficiency. This also allows us to deploy across regions, inside a customer’s cloud, or on-premises, serving this latency to clients throughout the world. Try it now in the API or in the playground with depth: "flash". Links in the comments.
-
-
Pruna AI builds fast, efficient AI models for image and video. Based in Munich and Paris, the company also maintains an open-source optimization framework featuring nearly 50 algorithms and a function called smash. That efficiency work is why P-Video-2, shipping today, has a draft mode. Video is billed per second of output, so the cost was never the clip you ship, it is the ten you deleted first. Now available all around the world through Eden AI. A German and French partnership that actually ships. 🇪🇺 Try it here: https://lnkd.in/etHQzj4h
Happy to announce P-Video-2: the fastest video generation model from text, images, and audio, with up to 20s videos, up to 1080p resolution, and 48 fps. • 𝗜𝗻𝗽𝘂𝘁: Text, image, or audio-conditioned generation • 𝗗𝘂𝗿𝗮𝘁𝗶𝗼𝗻: Up to 20s, or let the model choose the length from your prompt • 𝗤𝘂𝗮𝗹𝗶𝘁𝘆: Up to 1080p at 24 or 48 fps • 𝗦𝗽𝗲𝗲𝗱: Draft from ~0.41 s/s · Standard from ~0.91 s/s • 𝗣𝗿𝗶𝗰𝗶𝗻𝗴: Draft from $0.015/s · standard from $0.025/s • 𝗗𝗿𝗮𝗳𝘁 𝗺𝗼𝗱𝗲: Faster previews at half the standard price • 𝗖𝗼𝗻𝘁𝗿𝗼𝗹: Optional last-frame image, seed, aspect ratio, and prompt upsampling Text, image, or audio in. Video out. Available via our inference partners Replicate, Cloudflare, Runpod, Segmind, each::labs, WaveSpeedAI, Runware, Eden AI Validated by our benchmark partners Rapidata, Datapoint AI, Intelligence (Design Arena). Benchmarks to be released soon 👀 👉 Free Playground: https://buff.ly/090nICI 📚 Model docs: https://buff.ly/JHxxCNn 🧩 API: https://buff.ly/405NzxE
-
Eden AI a republié ceci
🇪🇸 Europe's highest-scoring AI model is Spanish, and you can call it today. Quasar 438B, from Multiverse Computing, scores 43 on the Artificial Analysis Intelligence Index. Highest of any European model. 13th out of 178 worldwide. 🔥 Here is the part we did not expect: it comes from a company that spent six years making other people's models smaller. This is the first model they built themselves. Live on Eden AI now: https://lnkd.in/ezeXVKCG Congrats to the Multiverse Computing team 🇪🇺 Full benchmarks and pricing in the comments 👇
-
-
🇪🇸 Europe's highest-scoring AI model is Spanish, and you can call it today. Quasar 438B, from Multiverse Computing, scores 43 on the Artificial Analysis Intelligence Index. Highest of any European model. 13th out of 178 worldwide. 🔥 Here is the part we did not expect: it comes from a company that spent six years making other people's models smaller. This is the first model they built themselves. Live on Eden AI now: https://lnkd.in/ezeXVKCG Congrats to the Multiverse Computing team 🇪🇺 Full benchmarks and pricing in the comments 👇
-
-
GLM-5.3-Flash just landed on Eden AI⚡ Z.ai’s new model brings together: → 320B total parameters, 18B active → Text, image and video understanding → 1,048,576-token context → $0.15/M input tokens → $0.50/M output tokens That combination makes it especially interesting for large-scale agents, document processing and multimodal workflows where model costs can quickly become the bottleneck. You can now test GLM-5.3-Flash alongside the rest of your AI stack directly through Eden AI without building another provider integration. 👉 Try it: https://lnkd.in/etHQzj4h
-
-
Developers can now access Firecrawl directly through Eden AI🔥 Scrape pages, crawl entire websites, run web searches, extract structured data, generate sitemaps, or launch deeper research workflows - all with your existing Eden AI key. 👉 More info: https://lnkd.in/eA5Ksg-H 👉 Try it: https://lnkd.in/etHQzj4h
-
-
We're a Day-0 launch partner for Resemble AI's DETECT-World, now available on Eden AI 🚀 Instead of memorizing attack signatures, DETECT-World evaluates whether lighting, motion and spatial consistency actually make sense. 👉 More info: https://lnkd.in/eBycpZFZ 👉 Try it, integrate and compare it with other providers: https://lnkd.in/etHQzj4h #resembleai #edenai #deepfakedetection #euaigateway
-
-
Congratulations Resemble AI team on this launch, amazing technique on deepfake detection 👏🎉
Today we're announcing DETECT-World 🌎 It's our third gen deepfake detection model, and the first one to be built using world model architecture. We've spent the last few years getting very good at teaching detection models what AI generated content looks like and training them quickly as new variants appear. But as the breadth and depth of realistic gen AI has exploded, that approach isn't enough. Hugging Face has gone from 2 million model variants to nearly 3 million in the last nine months, while we're seeing increasingly sophisticated deepfakes used for impersonation, identity fraud, and political disinformation. A lot of the time those deepfakes run on open source models. Deepfake detection has always been built around finding the traces that generative models leave behind. Now instead of only asking, “Have I seen this pattern before?” DETECT-World also asks, “Does this make sense in the real world?” It checks things like lighting, motion and timing. The idea is simple: a generator can be new to us, but it still has to produce something that makes sense in the real world. We benchmarked DETECT-World across 250+ generation models and saw 95.8% accuracy on image and 98.2% on video. Audio detection also improved this generation, now at 99.5% accuracy, externally validated by Podonos. Resemble DETECT-World is available today through our platform and via API. Try it in a real world scenario here > https://lnkd.in/gcuTG6VH Huge shoutout to Resemble AI's all-star research team for this breakthrough: Zhe Niu, Tedi Papajorgji, Zihan Jin, Nicolas Müller, John Meade, Aditya TB, Piotr Kawa, Matthew Hutter