MiniMax H3

MiniMax announced the launch of MiniMax H3 on July 31, 2026 as a general-purpose multimodal generation model that understands unified context across text, images, video, and audio, producing up to 15 seconds of video at 2K resolution with native stereo sound. The launch blog positions H3 for commercial content creation Technically, H3 is powered by Contextual Omni Representation, H3-VAE, H3-Omni Transformer, and In-Context Regeneration, with MiniMax claiming its 2K per-second price is less than a third of mainstream models and its 768p output is less than half the price of competitors' 720p. The post states MiniMax planned to release

Vercel's official AI Gateway pricing documentation states that the gateway charges no markup and no platform fee on tokens, with customers paying the provider's list price on a pay-as-you-go basis through purchased AI Gateway Credits. Every Vercel team account receives both a free tier ($5/month in credits, limited to The pricing page explains that the per-request charge depends on the selected AI provider and model plus input and output tokens processed, and that teams can cap spending at the team, project, API key, or member level via budgets. Free-tier requests are rate-limited per model and return 429 errors when exceeded, with

302.AI published a review on August 6, 2026 characterizing MiniMax H3 as a "full-modality video model" that combines generation, editing, and referencing into a single pipeline. It accepts mixed inputs of up to 9 images, 3 videos, and 3 audio clips at once, removing the traditional split between generation, editing, su The review highlights H3's text consistency improvements aimed at ending garbled text in generated video, as well as stable physical dynamics and state-of-the-art-level audio output. It frames H3's release alongside Seedance 2.5 as a catalyst that reactivated the AI video generation track after a quieter first half of

Kingy AI published a technical review on August 4, 2026 distinguishing the hosted MiniMax H3 pipeline (which includes H3-Context-IR and H3-Regenerate-2K stages not in the weight release) from the downloadable H3-Base, which produces 768p video only. The review notes the local BF16 base requires roughly 134 GiB of weigh Early independent evidence placed H3 in the top three across six Artificial Analysis preference pools and first for audio-inclusive video editing as of August 3, 2026, though these reflect the hosted pipeline rather than local renders. MiniMax had not yet published a model-quality benchmark table, multi-seed failure an

According to Enera's independent recap, MiniMax launched H3 on July 31, 2026 and open-sourced the weights on Hugging Face on August 3, 2026 under the MiniMax H3 Community License, making it the first open-weight model to reach the top of an independent AI video ranking. The piece reports Artificial Analysis placements Enera describes H3 as a 33-billion-parameter dense Transformer (not a mixture-of-experts) that ingests text, images, video, and audio as a single context from the start of training, and notes the system comprises three components in the model card, including H3-Context-IR for prompt interpretation. The article frames H