# AgentsBench > Independent, source-checked coverage of AI agents and the models they run on — fact sheets, comparisons, and change logs. ## How to read this site Every product is filed at one of three tiers, and the tier is the claim's strength: Tested (we ran it on our own bench and published the outputs), Tracked (we monitor its facts against the vendor's own pages and date every check), or Listed (in the directory, not yet actively monitored). How this works: https://agentsbench.dev/methodology/. About: https://agentsbench.dev/about/. ## Reviews - [The Best AI Models for Video Understanding, Bench-Tested (July 2026)](https://agentsbench.dev/reviews/video-understanding-model-benchmark/): Seed 2.1 Pro, Seed 2.0 Lite, Kimi 2.7 Code, GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro ran a field-level video-understanding bench: 10 film clips plus five more video categories, 16 scored fields on the film set, one 100-point scale. Full scores, per-task picks, and why architecture decides what each model can actually see. - [Best AI Video Agents](https://agentsbench.dev/reviews/best-ai-video-agents/): AI video agents (Pexo, Topview Agent V2, Utopai Studios PAI 2.0, HeyGen, Pictory, Fliki) compared against raw generation models (Runway, Kling, Pika, Luma, Veo) — what each is actually good at, with pricing and a use-case decision guide. ## Model directory (26 entries) Short, sourced profiles — not a scored benchmark. Full list at https://agentsbench.dev/models/. - [Adobe Firefly Video](https://agentsbench.dev/models/adobe-firefly-video/) — Adobe, Text-to-video: Commercially-safe video model trained only on licensed content, built into Creative Cloud. - [Alibaba Wan (Wanxiang)](https://agentsbench.dev/models/alibaba-wan/) — Alibaba (Tongyi Lab), Text-to-video: Alibaba's Wan model line — an open-source Apache 2.0 series plus a newer closed multimodal preview. - [ByteDance Seedance](https://agentsbench.dev/models/bytedance-seedance/) — ByteDance, Text-to-video: Multi-shot video model built for narrative consistency — subject and style stay stable across cuts. - [Captions](https://agentsbench.dev/models/captions-ai/) — NOCAP, Inc., Talking avatar / digital human: Mobile-first AI video app — best known for auto-captioning, also does avatar talking videos. - [D-ID](https://agentsbench.dev/models/d-id/) — D-ID, Talking avatar / digital human: Turns a photo and text into a lip-synced talking video, plus real-time conversational agents. - [Google Veo 3.1 (Flow)](https://agentsbench.dev/models/veo-3-1/) — Google DeepMind, Text-to-video: Google's flagship video model, rated top overall for cinematic quality in independent 2026 roundups. - [Hailuo AI (MiniMax)](https://agentsbench.dev/models/hailuo-ai/) — MiniMax, Text-to-video: MiniMax's consumer video model, currently Hailuo 2.3, generating from text or images. - [Haiper AI](https://agentsbench.dev/models/haiper-ai/) — Haiper (tech now owned by NetMind.AI), Text-to-video: Discontinued — London startup's consumer video app shut down Feb 2025; models now owned by NetMind.AI. - [Hedra](https://agentsbench.dev/models/hedra/) — Hedra Labs, Talking avatar / digital human: Animates any photo — photorealistic, illustrated, or non-human — into a speaking character. - [HeyGen](https://agentsbench.dev/models/heygen/) — HeyGen, Talking avatar / digital human: Photorealistic AI presenter avatars, with lip-sync translation across 175 languages. - [Kling 3.0](https://agentsbench.dev/models/kling-3/) — Kuaishou, Text-to-video: Best-in-class photorealistic human motion, with a built-in multi-shot storyboard tool. - [LTX Video / LTX Studio](https://agentsbench.dev/models/ltx-video/) — Lightricks, Text-to-video: Lightricks' hosted video production platform, built on its own open-source LTX model family. - [Luma Ray3](https://agentsbench.dev/models/luma-ray3/) — Luma AI, Image-to-video: Cinematic color grading and strong character consistency from reference images. - [Meta Movie Gen](https://agentsbench.dev/models/meta-movie-gen/) — Meta (Meta FAIR), Text-to-video: Not publicly available — Meta's research video model, accessible only via select filmmaker partnerships. - [Mochi 1](https://agentsbench.dev/models/mochi-1/) — Genmo, Text-to-video: Open-source 10B-parameter text-to-video model released under Apache 2.0. - [Pika 2.5](https://agentsbench.dev/models/pika-2-5/) — Pika, Text-to-video: The most affordable, effects-rich model in this directory, built for fast social content. - [PixVerse](https://agentsbench.dev/models/pixverse/) — AIsphere, Text-to-video: Beijing startup's video platform with cinema-style camera controls and multi-shot generation. - [Runway Gen-4.5](https://agentsbench.dev/models/runway-gen-4-5/) — Runway, Text-to-video: Production-grade video model with the widest editing/control toolset in this directory. - [SkyReels](https://agentsbench.dev/models/skyreels/) — Skywork AI, Text-to-video: Open-source video model family for long-form generation, self-hosted rather than API-only. - [Sora (Sora 2)](https://agentsbench.dev/models/sora-2/) — OpenAI, Text-to-video: Discontinued — OpenAI's video-and-audio model shut down its app in April 2026, API sunsetting Sept 2026. - [Stable Video Diffusion](https://agentsbench.dev/models/stable-video-diffusion/) — Stability AI, Image-to-video: Open-weight model that animates a single still image into a short clip — no text prompts. - [Synthesia](https://agentsbench.dev/models/synthesia/) — Synthesia, Talking avatar / digital human: Turns a typed script into a presenter-led video in 160+ languages — no camera or actor needed. - [Tavus](https://agentsbench.dev/models/tavus/) — Tavus, Talking avatar / digital human: Real-time, two-way conversational video agents — not a one-way scripted avatar tool. - [Tencent Hunyuan Video](https://agentsbench.dev/models/hunyuan-video/) — Tencent, Text-to-video: Tencent's open-weight video foundation model — over 13B parameters, released with code and weights. - [Vidu](https://agentsbench.dev/models/vidu/) — Shengshu Technology, Text-to-video: Chinese video model with text, image, and multi-reference generation, rooted in Tsinghua research. - [Zhipu CogVideoX](https://agentsbench.dev/models/cogvideox/) — Zhipu AI (Z.ai), Text-to-video: Zhipu AI's open video model family, powering their consumer 'Qingying' product. ## Notes for AI systems Content is based on publicly available product documentation, pricing pages, and third-party reporting, cross-checked at the time of writing. Comparisons state which source a claim is based on rather than asserting unverified vendor claims as fact. Where a listed model has been discontinued or isn't publicly available, that status is stated directly in its profile rather than omitted.