- ** Big bean bun model 1.6**
- Seedance 1.0 pro
- Core technology bright spot
- Billing and limit flow
- ** Beanbag voice podcast model**
♪ Come on, come on, come on, come on, come on, come on ♪
The bytes jumped under the flag of the volcano engine, released a series of new models, such as the large bean bag model 1.6, the video generation model Seedance 1.0 pro, and the bean bun voice podcast model, and upgraded AI cloud log services, such as the Agent development platform.
** Big bean bun model 1.6**
The soybean 1.6-thinking model has reached the level of the first global echelon on multiple authoritative scales: the GPQA Diamond test, with 81.5 points, is one of the best reasoning models available; the mathematical evaluation of AIME25, with 86.3 points, represents a significant increase of 12.3 points compared to the soybean 1.5 in-depth reflection model.
The large bean bun model 1.6 series supports in-depth thinking, multimodular understanding, 256k long context, graphical interface operations, etc., and is better able to support complex Agent construction and contribute to the enhancement of AI productivity.
Seedance 1.0 pro
Seedance 1.0 pro supports text and photo input, generating 1080p high-quality video with seamlessly transacted images, with a high degree of physical stability and image. On the internationally renowned rating list of Artificial Analysis, Seedance 1.0 pro is ranked first in both the graphic and graphic videos.
-
** Type of input**: Support text, images, videos, audio
-
** Output type**: generation of video (support for Text-to-Video and Image-to-Video)
-
Resolution support: 480 p, 1080 p
-
** Frame rate**: 24fps
-
** Length**: 5 or 10 seconds
Core technology bright spot
** ** Multi-coupled narrative ability**
-
Progenerate support for continuous multi-camera video generation
-
Adoption of multi-module location coding and unified modeling structures;
-
Achieving visual consistency and contextual consistency in time and space switching.
** High-real sports performance**
-
Fine data sets and incentive mechanisms based on self-research;
-
Optimizing the quality of production using RLHF technology;
-
To achieve high-security simulations from micro-expression to action scenes.
** Capability to follow instructions**
-
Precision of multi-subject, multi-action combinations;
-
High matching text description, controlling the behaviour of the “cargo mirror”;
-
Video missions in support of complex narrative structures.
** Multi-style video output** Supports the following style control words (prompt style):
- Paints, furs, watercolors, 3D cartoons, ink, clay, pixels, steam punks, sketches, etc.
Billing and limit flow
-
** Price **: $15 / million token (same price as graphics)
-
** Simultaneous issuance**: up to 10 requests per master account number
-
** Flow limit threshold**: up to 600 creation requests per minute (RPM)
Further technical details on Seadance 1.0 pro are available on the project home page (the full technical report can be viewed)
** Beanbag voice podcast model**
The newly released soybean voice podcast model, which is derived from the further expansion of end-to-end real-time voice, enables the generation of seconds from text creation to dual-dialogue podcast productions, with natural podcasting elements such as interconnection, interference, hesitation and the likeness of the real person.
