BOSS Zhipin’s Nanbeige releases a 33B model for coding and office agents
Nanbeige4.3-33B brings downloadable Apache-2.0 weights and a 512K context window. The lab plans to use it as a teacher for smaller agents.
THE GLOBAL DESK
Global.
Models and AI developers around the world, including Qwen, Kimi, MiniMax, Mistral and Ai2.
24 STORIESDATES REFER TO ORIGINAL ANNOUNCEMENTS
Kimi K3.1 Pro arrives with 1M context; weights promised by November 4
Moonshot’s new flagship and Highspeed variant are available through its products and API. The downloadable-weight release is still ahead.
MiniMax M3.1 Flash weights arrive for multimodal agent work
Official weights are now downloadable. The 428B-parameter model pairs a million-token context with efficiency changes, while commercial use carries license conditions.
Qwen Code makes running tasks easier to follow in its latest release
Version 0.25.3 adds live execution records and more workspace controls for people supervising coding agents.
Ai2 updates HiRO-ACE to simulate more kinds of extreme weather
The climate emulator adds detailed wind and pressure output alongside rainfall, with an important limit on climate generalization.
Kimi Code Desktop adds Linux support and remote control
The October update lets users steer a local agent from another device while execution remains on the host computer.
Z.ai opens beta applications for its GLM-Legal model
The specialized model targets legal work products, with API access by application and AutoClaw integration described as forthcoming.
Qwen Code adds live voice, browser work and email tasks
The October 9 update brings several ways to interact with the coding agent, with platform limits on Chrome integration.
Ai2 publishes new byte-level checkpoints as Bolmo reaches Nature
The research extends a conversion method beyond Olmo while giving researchers additional intermediate models.
Mistral Large 4 enters public preview ahead of a weights release
The European lab opens an API preview while scheduling downloadable weights for later in October.
Cohere updates North with agent orchestration and cost controls
North 2 combines workflow features with options for private and air-gapped deployment.
Ai2 opens AstaBrief, an 8B model for cited research reports
The release includes weights and training data, with a clear warning about the age of its evaluation comparisons.
Cohere releases Embed 5 with Pro and Fast retrieval models
Two tiers share an embedding space, allowing different query choices without rebuilding the same index.
Qwen Code improves context inspection and interrupted workflows
The September 24 roundup adds file-level context accounting, remote workspaces and selective retries.
Qwen-Image-2.1 unifies creation, editing and transparent output
The September open release uses a seven-billion-parameter visual generation component and accepts multiple references.
Qwen 3.8-Omni-Flash extends agents into audio and video work
The September announcement describes a million-token multimodal model and a runtime for real-time interaction.
Qwen LiveTranslate adds speaker separation to interpretation
The September model update combines bilingual text, context and translated speech in a streaming workflow.
Qwen Code adds external agents and limits for autonomous work
The September 17 release connects Claude Code and Codex while adding workflow caps, time limits and live web previews.
Mistral models begin powering Firefox Smart Window’s beta
The Mozilla partnership ties browser assistance to regional rollout plans and a stated zero-retention commitment.
Qwen Code gives workflows their own dashboard
The September 10 update makes agent runs inspectable and lets named group-chat tasks use separate workspaces.
Qwen-Drive-1.0 connects visual understanding to motion planning
The September research adds perception and planning modules to a general vision-language backbone.
MiniMax introduces Music 3.0 for full-song generation
The model combines structured musical descriptions with a redesigned generation and audio-rendering pipeline.
Qwen 3.8-Max launches with a focus on extended agent work
The August announcement scales the flagship to 2.4 trillion total parameters and promises open weights.
Qwen-Image-3.0 takes on dense layouts and small text
The July release targets information-heavy images such as newspapers, storyboards and illustrated documents.
↗
No Global stories in this edition yet.
This desk will feature source-linked coverage of other models and model teams around the world.
Browse all stories