Three Major AI Model Launches Reshape China's Competitive Landscape in July 2026

A widely circulated claim about a December 26, 2024 trifecta of Chinese AI model launches does not match available evidence. The three models in question — DeepSeek V4-Flash, MiniMax H3, and ByteDance Seedance 2.5 — all launched in mid-2026, not late 2024. Here is the corrected timeline, verified product specifications, and what these releases mean for the global AI landscape.

DeepSeek V4-Flash: April 24, 2026 Preview, July 31, 2026 Public Beta

DeepSeek V4-Flash first appeared on April 24, 2026 as the smaller half of the V4 family: a 284-billion-parameter Mixture-of-Experts (MoE) model with 13 billion active parameters per token . It shipped with a 1-million-token context window and MIT-licensed open weights .

API pricing settled at $0.14 per million input tokens and $0.28 per million output tokens, with cached input as low as $0.028 per million tokens . This makes V4-Flash one of the cheapest frontier-class reasoning APIs available — cheaper than Gemini 2.5 Flash at $0.30/$2.50 per million tokens .

A re-post-trained "0731 build" released on July 31, 2026 is the official public beta of the API. The 0731 build reportedly beats the larger V4-Pro-Preview on agent benchmarks, scoring 82.7 on Terminal Bench 2.1 and 76.7 on Cybergym . The model supports configurable reasoning modes (standard, high, and max thinking effort) alongside non-thinking mode for high-throughput pipelines .

MiniMax H3: July 31, 2026 — Open-Weight Omni-Modal Video

MiniMax launched H3 on July 31, 2026, positioning it as a general-purpose multimodal generation model capable of processing text, images, video, and audio in a unified context . The model generates up to 15 seconds of 2K-resolution (2560×1440) video at 24 fps with native stereo audio produced in the same inference pass — no separate audio pipeline required .

H3 accepts up to 9 images + 3 video clips + 3 audio tracks as simultaneous reference inputs . MiniMax committed to releasing the model weights within days, extending the open-weight approach into video generation where most commercial competitors remain proprietary . The company claims H3's per-second cost at 2K resolution is less than one-third of mainstream rivals .

ByteDance Seedance 2.5: June 23 Announcement, July 16 API Access

ByteDance announced Seedance 2.5 at its Volcano Engine FORCE conference on June 23, 2026 , with public API access opening on July 16, 2026 .

The headline upgrade: 30-second single-pass video generation — delivering a full clip in one continuous generation without stitching shorter segments together . Resolution reaches up to 4K (3840×2160) with 10-bit color depth . The model accepts up to 50 multimodal reference inputs — images, audio clips, 3D white models, style frames, and scene direction — up from 12 in the previous version .

ByteDance also released region-level editing, allowing users to modify specific areas of a generated clip without regenerating the entire sequence . Applications span film, advertising, education, industrial manufacturing, and autonomous driving simulation .

Strategic Implications of the July 2026 Wave

These three releases, clustered heavily on July 31, 2026, share common strategic themes:

• Aggressive pricing: DeepSeek V4-Flash undercuts Western reasoning models by 1–2 orders of magnitude . MiniMax H3 explicitly targets "less than one-third" the per-second cost of competitors .
• Open-weight momentum: Both DeepSeek V4-Flash (MIT license) and MiniMax H3 (promised weights within days) continue the Chinese AI industry's shift toward releasing model weights publicly — a move that pressures proprietary Western video generators .
• Longer context and multimodal integration: All three models push boundaries on input length (1M tokens for DeepSeek, 50 references for Seedance 2.5) and multimodal understanding (text/image/video/audio in a single pipeline for MiniMax H3).

Caveats and Unverified Claims

The original framing suggested these three launches occurred on December 26, 2024, alongside broader claims about Chinese open-source models representing 41% of global downloads and U.S. allegations of model distillation. The available search results do not support either the December 2024 date or those specific statistics. Those claims may relate to earlier DeepSeek-V3 or Qwen events, but this analysis could not verify them through the provided sources.

Corrected Timeline Summary

Model	Launch Date	Key Specs
DeepSeek V4-Flash	Preview: Apr 24, 2026; Public beta: Jul 31, 2026	284B MoE (13B active), 1M context, $0.14/M input tokens
MiniMax H3	Jul 31, 2026	15s 2K video, native stereo audio, open weights promised
ByteDance Seedance 2.5	Announced Jun 23, 2026; API Jul 16, 2026	30s single-pass, 4K, 50 multimodal reference inputs

These three models — DeepSeek V4-Flash, MiniMax H3, and ByteDance Seedance 2.5 — represent a coordinated acceleration of China's AI sector in mid-2026, defined by lower cost, open-weight distribution, and dramatically longer generation capabilities.