Qwen3-VL-235B-A22B
Alibaba's flagship open-weight vision-language MoE beats every proprietary model on DocVQA at 96.5% and MathVista at 85.8%, but trails GPT-5.4 and Gemini 3.1 Pro on broad MMMU-Pro reasoning.
Models
DeepSeek-VL2
DeepSeek-VL2 is DeepSeek's open-weight Mixture-of-Experts vision-language model, activating just 4.5B of its 27B parameters to hit 93.3% on DocVQA and beat GPT-4o on OCRBench.
Qwen2.5-VL-72B-Instruct
Alibaba's dense 72B vision-language model tops the open-weight DocVQA leaderboard at 96.4% and remains the default self-hosted choice for document and chart understanding.
AlayaWorld
AlayaWorld is a 15B open-weight video diffusion world model from Alaya Lab that sustains interactive, camera-controllable environments past 60 seconds.
Gemini 3.6 Flash
Google DeepMind's workhorse Flash model cuts output tokens 17% versus Gemini 3.5 Flash, drops output pricing to $7.50/M, and cuts DeepSWE task tokens by 65% while trailing GPT-5.6 Luna and Grok 4.5 on raw coding scores.
DeepSeek-R1
DeepSeek-R1 is the 671B-parameter open-weight reasoning model that matched OpenAI o1 on math and coding benchmarks and triggered a $589 billion single-day drop in Nvidia's market cap in January 2025.
Luma Ray3.2
Luma Ray3.2 is Luma AI's current flagship video model - native 16-bit HDR, 16-keyframe control, and the company's first full developer API, but still no native audio.
Pika 2.5
Pika Labs' flagship video model trades cinematic Elo rankings for the deepest creative-effects toolkit in AI video, plus a pivot into real-time agent video with PikaStream.
Haiper 2.x
Haiper 2.x is the cheapest per-second AI video API on the market at $0.033/sec, now run by NetMind.AI after Haiper's consumer app shut down and its founders joined Microsoft.
Hailuo 02
MiniMax's Hailuo 02 pairs a physics-aware architecture with some of the cheapest per-second video pricing in the industry, though newer rivals have since passed it on raw quality.
Mochi 1
Genmo's Apache 2.0 licensed 10B-parameter video generator is the largest open-weight text-to-video model released, with no managed API and roughly $0.33 per clip to self-host on an H100.
Qwen3.8-Max-Preview
Alibaba's 2.4 trillion parameter multimodal MoE model claims to trail only Claude Fable 5, but ships with no model card, no benchmark table, and no confirmed pricing.
