210 results. Find your next favorite, with a little help from the community.
Foundation models you can call or deploy
ModelTencent's open-source 13B parameter text-to-video foundation model. HunyuanVideo-1.5 upgrade brings higher quality and efficiency for video generation.
AI21's hybrid SSM-Transformer architecture model with 256K context window, optimized for enterprise RAG and long-context document processing tasks.
Moonshot AI's code-specialized model from the Kimi K2 family, optimized for software engineering tasks with strong benchmark performance.
Kuaishou's breakthrough AI video model with native 4K output, multi-shot storyboarding, lip-synced bilingual audio, and 15-second cinematic clips.
Xiaomi's reasoning model scoring 53.83 on Intelligence Index, demonstrating strong capability in math and code from a consumer electronics giant.
Unified MoE model combining reasoning, vision, and coding. 119B total/6B active params, 256K context, configurable reasoning depth. Apache 2.0.
Meta Superintelligence Labs' flagship model replacing the Llama line. 1M token context, 131K max output. Ranked #1 on TaxEval v2 out of 123 models.
NVIDIA's 120B open-weight MoE reasoning model (12B active) with 1M context and 350 tok/s output speed. Open weights on HuggingFace.
NVIDIA's 550B parameter MoE model with 55B active parameters, delivering frontier-level performance for enterprise AI deployments and inference optimization.
Poolside AI's coding-focused foundation models (M.1 flagship 225B MoE and XS.2 open-weight) built specifically for agentic software engineering.
Alibaba's Qwen 3.6 family released in six variants including Max Preview, scoring 51.81 on the Intelligence Index. Open-weight with strong multilingual support.
Alibaba's efficient 35B MoE model (3B active) ranked #1 on six coding/agent benchmarks including SWE-bench Pro and Terminal-Bench 2.0.
ByteDance's frontier language model released June 2026, part of their Seed foundation model series powering Doubao and internal AI products.
ByteDance's AI video generation model producing native 30-second 4K clips with up to 50 multimodal reference inputs. Strong temporal coherence and character consistency.
Google's prompt-to-video generation model with camera movement, lighting, dialogue, and lip sync capabilities integrated into Gemini.
OpenAI's optimized 809M parameter speech recognition model with 4-6x faster inference for voice agents while maintaining near-full accuracy.
01.AI's 100B parameter MoE model ranked 6th globally and 1st in China on LMSYS Arena, optimized for fast inference with strong reasoning capabilities.
OpenAI's first reasoning model, trained via reinforcement learning to think step-by-step before answering; targets math, science, and coding benchmarks over GPT-4o.