EASYHUB JOURNAL / 2026-08
AI news in August 2026
29 stories, ordered by their original dates. Open a title for its summary, image and source link.
- Ollama
Ollama’s transparent pricing
Ollama explains included usage and per-token billing. Current plan terms remain available on its official site.
- Tencent
Tencent Releases and Open-Sources Tencent Hy4 preview
Hy4 preview expands Hunyuan to 770B total parameters, 49B active parameters and million-token context for coding, cross-document office work and research. Open weights and access through WorkBuddy, CodeBuddy, Yuanbao, ima, TokenHub and OpenRouter are announced. This remains a preview, and launch-time trials are not permanent free access.
- Anthropic
Expanding our support for scientists
Anthropic outlines expanded subscriptions and research-support programs, including eligibility and scope.
- Cohere
Introducing Parse: Enterprise document intelligence at scale
Cohere introduces Parse to turn complex documents, tables and images into structured content for search and agent workflows, including Markdown output. The announcement describes multiple deployment channels, with cost and quality results tied to the publisher's stated evaluation conditions.
- Google
Gemini Omni 1.1 Flash lets you build with more control
Gemini Omni 1.1 Flash adds scene extension, first-and-last-frame controls, 4K upscaling and lighter draft generation for iterative video workflows. Upscaling is a post-processing capability and should not be described as native 4K generation for every output.
- Anthropic
Previewing the Model Hardware Standard
Anthropic opens a research preview of the Model Hardware Standard to selected laboratories and manufacturers, exploring common interfaces for agents and instruments. It remains a partner preview, not a universally deployed or finalized open standard.
- Alibaba Cloud
Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiency
The Qwen team opens Qwen3.8-Flash-Next weights and describes changes to attention, residuals, embeddings and optimization for multimodal and tool-driven work. It previews architectural ideas for the next generation, not the launch of the full Qwen4 family.
- Google
Intelligent transcription with Gemini 3.5 Transcribe
Google introduces Gemini 3.5 Transcribe for real-time transcription, captions and voice applications, including handling noise, specialized vocabulary and disfluencies. Accuracy still depends on recording quality, language and the intended setting.
- Alibaba Cloud
Alibaba Launches QwenWork International Edition, Extending Its All-in-One Workplace AI Agent to Global Markets
QwenWork's international edition enters public beta on web and desktop, combining cross-app tasks, web building and multimodal creation. Users authorize multi-step work and can save successful workflows as reusable skills.
- Z.ai / 智谱
GLM-5.3-Flash: Frontier Intelligence, Flash Cost
Z.ai releases GLM-5.3-Flash with a newly trained base, native multimodality and a hybrid attention architecture aimed at efficient inference. Weights and local-serving integrations accompany the release. It is not simply a cheaper alias of text-only GLM-5.3, and launch discounts are temporary.
- Ollama
Claude Desktop support with Ollama
Ollama describes connecting Claude Desktop through a third-party gateway for open-model workflows.
- Sarvam AI
Introducing Saaras V4
Sarvam AI introduces Saaras V4 and details its speech encoder, in-house language model, output modes and evaluation methodology. The release emphasizes Indian languages, English and difficult real-world audio. Performance should be assessed for the actual language and conditions of use.
- Adobe
Adobe Firefly expands its creative AI studio: generate music, speech, and sound effects in one place
Adobe broadens Firefly music, speech and sound-effect generation, with updates to its creative assistant and partner models. Output terms must be checked separately; the product's licensing claims do not automatically grant republication rights to images in the announcement.
- Alibaba Cloud
Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model
Alibaba Cloud summarizes the release of Qwen3.8-27B and flagship model weights. The smaller model targets multimodal work and lighter deployments, while the flagship serves larger setups. This entry uses the announcement date, not an inferred first repository-upload time.
- Anthropic
How Claude’s text watermark works
Anthropic explains text watermarking planned for future Claude models, including output quality, probabilistic detection and privacy. The post describes a plan and mechanism, not proof that every current Claude response already contains the watermark.
- Z.ai / 智谱
GLM-5.3: Frontier Coding with Emergent Cyber Capabilities
Z.ai introduces GLM-5.3, building on the GLM-5.2 base through expanded post-training for complex coding and longer tasks. The announcement discusses safety testing and schedules a later weight release. Its reported evaluations are publisher claims rather than independently verified rankings.
- Cursor
Cloud agents start 3x faster with builds
Cursor Builds prepares repositories, dependencies and completed setup scripts in advance for cloud agents. Sessions can reuse the last successful environment when a new build fails. The speedup in the original title is publisher-reported, not guaranteed for every repository.
- DeepSeek
DeepSeek V4 Pro reaches general availability
DeepSeek announces the production V4 Pro release across its app, website and API, alongside agent-oriented improvements and Responses API compatibility. The pricing announcement has a separate effective date; this archive entry uses the announcement date.
- Google
Introducing Gemini 3.7 Flash
Google introduces Gemini 3.7 Flash with improvements aimed at software engineering, knowledge work and web development. The announcement discusses capability and cost trade-offs; introductory prices and evaluation results reflect conditions at publication.
- MiniMax
MiniMax Music 3.0 advances full-song generation
MiniMax introduces Music 3.0 to compose, arrange and produce complete songs from a creative prompt and optional lyrics. Its technical article explains the handling of song structure, instrumentation and vocals. Quality claims describe the publisher's results, not an independent reproduction.
- IT之家
WorkBuddy synchronizes desktop tasks with mobile approvals and stopping
Tasks, conversations and outputs now synchronize across PC, mobile apps and the mini program. Mobile users can approve or stop desktop tasks and browse workspaces. Connected-computer mode requires the computer to remain powered on and both devices to use the same account, with app 1.2.0 and PC 5.3.8 or later.
- Ollama
NVIDIA Nemotron 3.5 Lightning
Ollama adds NVIDIA Nemotron 3.5 Lightning for local agents that gather context, call tools and work through multiple steps. The post also describes an MLX option for Apple Silicon; practical deployment still depends on model size and available memory.
- Sarvam AI
Indic DiarBench: A Joint Diarization-ASR Benchmark Dataset for Indian Languages
Sarvam AI publishes an open benchmark combining speech recognition and speaker attribution across Indian languages. Audio, annotations and an evaluation protocol cover multi-speaker conditions, including interruptions and overlap, rather than relying only on single-speaker transcription tasks.
- Ollama
Muse Glimmer from Meta Superintelligence Labs is now available
Ollama adds Meta's open Muse Glimmer model for local coding and longer-running agents, including image input. The release describes MLX acceleration and execution options on Apple Silicon for developers exploring local-model workflows.
- Cursor
How Cursor Router chooses the right model for the task
Cursor explains how Auto Intelligence and Auto Balance select models using task characteristics and production feedback. The article describes a previously launched router and subsequent improvements, so its publication is an explanation update rather than the router's original launch.
- ByteDance Seed
SeedRealtime introduces audiovisual full-duplex interaction
ByteDance Seed introduces SeedRealtime, combining streaming audio, video and text for full-duplex interaction and contextual assistance. The announcement discusses its use in Doubao and publisher evaluations; model capability should not be confused with unrestricted access to devices or services.
- Tencent
Tencent Hy3 Now Available Globally, Extending Practical AI Across Products, Workflows and Cloud Services
Tencent broadens international access to the Hy3 model released in July through WorkBuddy, creative studio Miora and TokenHub. Miora targets graphics, video, 3D and interface assets, while TokenHub provides multi-model API management and routing. The change is product integration and distribution, not another Hy3 model.
- MiniMax
Open General Intelligence: MiniMax H3 Is Now Open Source
MiniMax announces the open release of H3, with checkpoints and supporting resources for video-generation tasks. The model repositories specify supported workloads and deployment requirements. Releasing model assets is distinct from making every hosted service free.
- Alibaba Cloud
Alibaba Unveils Qwen3.8-Max: Its Largest and Most Capable Flagship Model to Date
Alibaba Cloud introduces Qwen3.8-Max for multimodal understanding, coding, research and longer tasks through specified APIs and products. Model weights are described as a forthcoming release in this announcement, not as already downloadable on that date.