B.AI October Ecosystem Monthly Report | Full Expansion of the Multimodal Flagship Matrix, New Additions to the Large Model Leaderboard Witness the Comprehensive Rise of Chinese Large Models

In October, B.AI reached another major milestone in its platform development. Registered users surpassed 2,920,008, approaching the 3 million mark. The platform brings together more than 70 of the world's top multimodal large models, accessible through dual entry points via API and Web Chat. As both user scale and model supply climb in tandem, economies of scale are continuing to manifest on the platform. Behind this growth is the choice made by global developers and users in favor of B.AI's unified entry point and stable services.

On the model supply side, B.AI deepened its push into multimodality this month. Four video generation models from ByteDance's Seedance series launched at once, extending the platform's capabilities into video creation with output quality of up to 4K. At the same time, new offerings including GPT-6.1 Terra, Claude Haiku 5.5, MiniMax M3.1 Flash, and Kimi K3.1 Pro arrived one after another, further enriching the B.AI large model matrix. On the experience side, B.AI ensures that model iterations no longer become a burden for developers. The MiMo-V2.5 series achieves smooth migration through automatic routing of legacy model IDs, while DeepSeek V4.1 Flash can be upgraded to version 1017 without changing the model ID—businesses can access new capabilities without any changes. Meanwhile, three rounds of discount benefits for popular models rolled out in succession, with discounts as low as 90% off, making high-performance AI calls more cost-effective.

The richness of model supply and the upgrade of user experience are ultimately reflected in developers' real choices. In October, B.AI's all-new large model leaderboard officially launched, presenting the overall rankings and trends of major models on the platform based on real user call data. The rankings show that Chinese large models demonstrate dominant ecosystem appeal, sweeping the top eight spots on this month's call ranking and igniting the battle for developer traffic. Among them, DeepSeek-V4.1-Flash holds the top spot, while MiMo-V2.6-Flash emerges as this month's core growth engine with strong upward momentum.

Along with the leap in ecosystem scale and the continuous evolution of the technological foundation, here are the key developments for B.AI in October across the dimensions of the large model landscape, multimodal infrastructure, and developer benefits:

1. New Large Model Leaderboard: Seeing the Model Landscape Through Real Calls

Competition in the large model赛道 has entered deep waters, and developers' real call data has become an important standard for testing a model's comprehensive strength. In October, the B.AI platform released a new version of its large model leaderboard based on massive daily real Token consumption and API call frequency. The most core highlight of this month's rankings is undoubtedly the full-scale breakout of Chinese large models. They have not only built significant advantages in cost control but also achieved leapfrog catch-up in call popularity and frontier intelligence levels, becoming the mainstay of the global developer ecosystem. From popularity and cost to intelligence, the October rankings present the following core trends:

1. Peak Popularity: Chinese Large Models Sweep the Top 8 Call Rankings, Igniting the Developer Traffic Battle

In the "call ranking" that represents absolute influence, Chinese large models have not only established an overall advantage by sweeping the top eight, but competition among the leading models is also highly intense: leaders continue to consolidate their traffic base, while challengers rapidly close in with high growth rates. The positioning battle among major vendors around high-frequency business scenarios has fully unfolded.

• Absolute Dominance: DeepSeek-V4.1-Flash firmly holds the top spot on both call ranking lists (Token usage and request count). During the period from October 26 to November 1, its weekly Token usage share reached 81.1%, building an extremely deep traffic moat within the platform.
• Growth Engine: Close behind is Xiaomi's ecosystem model MiMo-V2.6-Flash, whose call popularity shows a strong upward trend and is mounting a challenge for the top spot.
• Second Tier in Dense Pursuit: Following closely are GLM-5.3-Flash, Qwen3.8-Flash, MiMo-V2.6-Pro, DeepSeek-V4-Flash, MiniMax-M3.1-Flash, and DeepSeek-V4-Pro, forming a strong chasing tier. Relying on differentiated task performance and highly competitive response performance, the second-tier camp is rapidly capturing traffic from segmented scenarios.

2. Diversified Cost Advantages, Empowering Developers to Reduce Costs Across All Scenarios

The "price ranking" reflects the real competition of global large models in commercialization. From October data, international and Chinese leading models have formed a differentiated competitive landscape across different billing dimensions. At the same time, cost competition among Chinese large models has entered a deep-water zone of "fighting over every cent," driving the release of extreme cost-effectiveness.

• International Models Solidify the "Comprehensive and Input" Benchmark: GPT-5 nano strongly tops both the weighted price and input price rankings (from low to high). Meanwhile, Claude Haiku 5.5 and GPT-6 Luna also firmly hold top five positions on the weighted price ranking and top four on the input price ranking. This indicates that in building massive knowledge bases, conducting ultra-long document reading (RAG), and general intelligent dialogue scenarios, international models remain highly competitive computing foundations.
• Chinese Large Models Dominate "High-Concurrency Generation" Scenarios: In the "output generation" segment, Chinese large models have established a significant structural advantage. MiMo-V2.6-Flash, MiniMax-M3.1-Flash, GLM-5.3-Flash, and Qwen3.8-Flash strongly sweep the top four lowest output prices on the platform. Comparing the internal pricing of these four leading domestic models, one can intuitively see the degree of refinement in competition within this range. Second-ranked MiniMax ($0.3 per million Tokens) has only a tiny price gap of $0.02 from the top; GLM ($0.35 per million Tokens) and Qwen ($0.376 per million Tokens) also closely pursue with cent-level price gaps.

3. Strong Breakthrough in Intelligence Level: Chinese Large Models Enter the Global First Tier

Chinese large models do not win solely on cost-effectiveness; in the "intelligence ranking" that evaluates logical reasoning and complex task capabilities, they have also delivered impressive results.

• Strong Breakthrough: Kimi K3.1 Pro, launched by Moonshot, with a high score of 55.1, strongly entered the top three of the global intelligence ranking (second only to Claude Opus 5.5 and Claude Sonnet 5.5).
• Blooming in Multiple Spots: Alibaba Cloud's Qwen3.8-Max, with a score of 53.4, firmly holds fifth place globally. On the most cutting-edge intelligence arena, Chinese large models have firmly taken a seat at the global first-tier table.

* For more ranking details and multi-dimensional leaderboard data, please visit B.AI Large Model Leaderboard.

Overall, platform developers have completely moved beyond single price considerations in computing power selection, shifting toward refined on-demand scheduling. Whether it is the "cost-effectiveness engine" supporting high-frequency concurrency or the "high-intelligence brain" driving complex reasoning, large models with different positioning can precisely lock in their own value niche within B.AI's real business flows.

2. Core Data and Platform Milestones: Economies of Scale Continue to Amplify

In October, B.AI achieved new progress across dimensions including user scale, model ecosystem, and multilingual coverage. The platform's economies of scale are further emerging, steadily moving from "fully featured" toward "mature at scale."

• User scale continues to expand: registered users reached 2,920,008, surpassing 2.9 million and approaching the 3 million threshold;
• Model ecosystem further enriched: more than 70 top large models are integrated into the unified entry point, with model aggregation capabilities continuing to strengthen;
• Key breakthrough in globalization capabilities: B.AI's official website, Web Chat, and the newly launched Android App v1.3.0 now fully support 11 major international languages. The comprehensive performance upgrade of the mobile端 and the breaking of language barriers not only bring a smoother native-language-level experience but also comprehensively lower the access threshold for global developers and users.

3. Top Model Matrix Expansion: Doubling Down on Multimodal Infrastructure to Build Full-Scenario Computing Power Supply

Behind the scale growth is continuously enriched model supply. This month, B.AI continued to expand its premium model matrix and focused on building "multimodal" infrastructure. From native visual understanding to joint audio-visual generation, we are committed to enabling developers to seamlessly call full-dimension top-tier models for processing text, images, and audio-video on a single platform, precisely meeting complex business needs. To help developers more intuitively select the right computing power, the heavyweight models newly integrated this month achieve full-scenario coverage in positioning and capabilities.

1. Seedance Series (ByteDance), Official Group API Now Fully Integrated

• Model Positioning: Multimodal audio-visual joint generation flagship matrix
• Core Highlights: Includes four refined models—Seedance 2.5, Seedance 2.0, Seedance 2.0 Fast, and Seedance 2.0 Mini. The 2.5/2.0 flagship versions focus on high image quality and strong control (up to 4K resolution, up to 30-second generation, support for up to 50 materials and first-last frame animation control); the Fast/Mini versions balance generation speed and extreme cost-effectiveness (480p/720p).
• Applicable Scenarios: Prompt trial-and-error, rapid iteration of advertising storyboards, short videos, and professional film-grade creation.

2. Kimi K3.1 Pro Series (Moonshot), Available on Both API and Web Chat

• Model Positioning: Ultra-long context and native visual understanding flagship
• Core Highlights: Features 1M Token ultra-long processing capability and powerful native visual parsing. Available in standard and Highspeed versions (2-3x faster response).
• Applicable Scenarios: Complex programming and code refactoring, CAD/3D modeling parsing, long-form financial research report analysis, and latency-sensitive multi-step Agent workflows.

3. MiniMax M3.1 Flash (MiniMax), Available on Both API and Web Chat

• Model Positioning: Complex logical reasoning and ultra-large output powerhouse
• Core Highlights: Combines powerful visual reasoning capabilities, supports 1M ultra-long context input, and provides an output Token cap of up to 512K.
• Applicable Scenarios: Complex office automation execution, software engineering and front-end code implementation, and long-form presentation creation.

4. Claude Haiku 5.5 (Anthropic), Available on Both API and Web Chat

• Model Positioning: Ultra-fast lightweight text and vision model
• Core Highlights: While maintaining extremely low latency and high cost-effectiveness, it offers 1M Token context and five adjustable levels of thinking depth, with notable improvements over the previous generation in programming, computer operation, and knowledge work.
• Applicable Scenarios: High-concurrency, latency-sensitive classification, routing, information extraction, and sub-agent tasks, as well as real-time streaming responses.

5. GPT-6.1 Terra (OpenAI), Available on Both API and Web Chat

• Model Positioning: Efficient visual reasoning and professional text creation model
• Core Highlights: Combines adjustable reasoning capabilities with a 1.05M ultra-long context window, supports up to 128K output Tokens, and while balancing strong performance (standard pricing $1.00/M input, $5.00/M output), provides flexible control for complex applications.
• Applicable Scenarios: Precise computer operation instruction generation, clear and natural professional writing, and daily high-frequency visual reasoning workflows.

* Log in to B.AI to unlock full-dimension top-tier large models and start your multimodal intelligent experience now.

4. Technological Experience Leap: Silky Integration, Seamless Switching

As the number of models continues to increase and iteration pace accelerates, the core challenge developers face has shifted from "insufficient selectable models" to "excessive follow-up costs." In response, B.AI is working simultaneously on integration, migration, and upgrade processes to continuously reduce the burden model updates place on businesses.

1. Smooth Migration: MiMo-V2.5 Series Automatic Routing Ensures Business Continuity

Facing the migration of the MiMo-V2.5 series to next-generation models, B.AI has introduced a smooth transition solution: the platform relies on its underlying scheduling capabilities to implement automatic routing for legacy model IDs, while matching new billing standards simultaneously. At the same time, sufficient code adaptation and testing windows are reserved for developers, ensuring that all types of application services remain uninterrupted before the old IDs are officially discontinued.

2. Upgrade Without Changing ID: DeepSeek V4.1 Flash Upgraded to Version 1017

The same applies to the DeepSeek V4.1 Flash upgrade. After upgrading to the latest version 1017, users can directly experience new capabilities without changing the model ID: performance across 11 benchmark tests is comprehensively improved, with DeepSWE v1.1 achieving a top-tier pass rate of 77.0%; responses are faster and outputs more refined, with shorter time consumption for similar tasks; Agent and programming capabilities are also optimized. Users can obtain stronger performance and higher efficiency while maintaining their existing integration methods.

3. B.AI Web Chat Adds Model Search Function

Facing the selection cost brought by 70+ models, B.AI Web Chat has added a model search function. Simply enter a name in the model selection window to quickly locate the target model—from massive choices to precise direct access, switching is more efficient and use is more convenient.

5. Inclusive Computing Power and Developer Benefits: Making High-Performance AI Calls More Cost-Effective

In October, B.AI launched multiple consecutive rounds of discounts around popular large