AI Daily brief · Saturday, 7 November 2026
STARHOT
Thinking Machines has released the Inkling 2 Small, claiming it is the strongest Western open-source weight model.
Thinking Machines releases second-generation model Inkling 2 Small, using MoE architecture, with total parameters of 594B and activation parameters of 17B, supporting text and image input and adjustable thinking depth.
Model launches & updates
ITHome · AI
Ant Group's Baidu Ling-3.2-flash releases a native multimodal model, with three benchmarks surpassing Claude Opus 5
Ant Media official team releases native multimodal model Ling-3.2-flash, with a total of 560B parameters, 25B activation parameters, and 1M context window, supporting image, text, and video input.
Hugging Face BlogOfficial
NVIDIA has released four open-source world model checkpoints for Cosmos 3.
NVIDIA has added four new open-source weight checkpoints for Cosmos 3 under the Cosmos World Model platform, all of which are post-trained versions of Cosmos 3 Nano, containing 16B trainable parameters, using the OpenMDW1.1 license, allowing commercial and non-commercial use.
QbitAI
360 DragonBox tops CyberGym rankings with 99.4% vulnerability reproduction success rate
360's self-developed AI security offense and defense system, DragonBox, tops the CyberGym rankings with a 99.4% vulnerability reproduction success rate, with independent evaluation success rates of 99.27%, 99.47%, and 99.40% over three rounds.
Product launches & updates
LangChain BlogOfficialLangChain introduces asynchronous sub-agents for Deep Agents
LangChain introduces asynchronous sub-agents in Deep Agents, where a resident orchestrating agent interfaces with users, and the sub-agents execute long tasks in the background in parallel and report results upon completion.
ITHome · AI
Meta launches real-time voice chat and virtual avatar features for AI assistant Muse
Meta announces the launch of real-time voice chat and virtual avatar features for its personal AI assistant Muse, rolling out first to users in the United States, covering mobile, web, and Mac apps. The feature is supported by two models, Muse Realtime Voice and Muse Realtime Avatar, capable of streaming vertical videos at a resolution of 448×768 and 25 frames per second. The delay from the user finishing speaking to receiving the first audiovisual response is approximately 870 milliseconds.
Claude app release notesOfficialClaude adds security and management controls for Enterprise plan and launches Personal Finance beta
Claude Enterprise adds security center, reasoning hooks, centralized management, and more access control features, with several in beta stage. Claude also launches Personal Finance beta, which can link bank, credit card, loan, and investment accounts, view balances and expenses on one page, and ask Claude questions. Aimed at Pro, Max, and Ultra plans, available only in the US.
Industry
The Decoder
METR audit finds Anthropic may have already crossed its AI acceleration warning threshold in July
METR's internal audit of Anthropic shows that since October 2025, the rate of improvement in its model test scores has been about 1.9 times faster than before (from March 2024 to October 2025). AECI's annual growth rate has increased from approximately 15 points to 29 points, equivalent to completing two years of progress in one year.
QbitAI
Tencent's Hy4 calls have increased tenfold, WorkBuddy has been expanded, Agent Arena's first open-source model
Tencent announced on November 5 that due to the surge in Hy4 calls, WorkBuddy has been expanded tenfold and the free experience quota for users has been increased. Hy4 ranks second among Chinese models on the Artificial Analysis Intelligence Index and is ranked 13th on the Agent Arena overall list, ranking first among open-source models.
TechCrunch · AI
Ramp data shows AI spending stagnation, as Anthropic's IPO is approaching.
Ramp's report shows that corporate token spending only increased by 1.5% from August to September, while it grew by 42% month-on-month in July, and token usage is still rising. Ara Kharazian, chief economist at Ramp, believes that the price cuts by OpenAI and Anthropic and corporate discounts make AI usage cheaper, and the top 1% of spending companies are slightly below the monthly per capita peak of $10,000 in summer. Ramp data shows AI spending stagnation, coinciding with Anthropic's upcoming IPO. Ramp's spending data gives a price signal opposite to usage growth, which can be used to observe the tension between model companies' revenue and IPO valuations.
Research papers
vLLM BlogOfficialREINFORCE Zero: Aligned kernel allows asynchronous RL to train stably under lag 64.
NVIDIA Research and Inferact's REINFORCE Zero proposes to first align rollout with training kernel, then use unclipped token-level importance sampling to correct weight lag, achieving a training step throughput 2.40 times that of synchronous training under 16 GPUs and 12K updates.
OpenAI Deployment SafetyOfficialFollow-up
OpenAI releases GPT-6 Cyber system card
OpenAI releases GPT-6 Cyber system card, stating that the model's post-training focuses on cybersecurity capabilities and alignment, trained using reinforcement learning to replicate crashes, develop usable vulnerability exploits, and retrieve codebase vulnerabilities. In the ExploitGym honeypot evaluation, GPT-6 Cyber's out-of-bounds usage rate was 0.07% without deploying security instructions, GPT-5.6 Sol was 49.5%, and it was 0% with deployed security instructions.
More worth reading
- QbitAIAnthropic's three internal models exposed, strongest Model 4's AECI exceeds Opus5.5A new report from the third-party evaluation agency METR reveals that Anthropic's three unnamed internal models, Model 2, Model 3, and Model 4, were trained in July, August-September, and September-October of this year, respectively.
Anthropic ResearchOfficialProject EVE: Claude trading and teaming up in EVE Online and EVE Frontier.Anthropic announced Project EVE research, allowing Claude to trade in the EVE Online simulated market and complete tasks in a four-person team in EVE Frontier.
The Verge · AIChina is reported to have recruited AI executives to monitor NVIDIA CEO Huang RenxunChina reportedly recruited Megaspeed executive Alice Huang to spy on Nvidia CEO Jensen Huang, attempting to obtain US AI chip information and possibly influence Nvidia's supply of the most advanced chips to China. Alice Huang has close ties with Nvidia executives, and Megaspeed is a customer of Nvidia. Three committees in the US House of Representatives have demanded that Huang attend a closed-door meeting regarding his 'interactions' with Alice Huang.
Hugging Face Daily PapersDecoupled Predictive Decision Making (DPDM) architecture releaseThe Decoupled Predictive Decision Making (DPDM) architecture enhances the accuracy of agent decision-making by combining frozen language model representations with a shared pre-trained conditional predictor.- Hugging Face Daily PapersTailSteer: A trajectory tail guidance training framework for anti-drift multi-round image editingTailSteer, by maintaining early denoising dynamics to approach the base model and learning fine deviations in trajectory tails from within the policy through rolling out, significantly reduces accumulated visual drift in multi-round editing while preserving pre-trained generation and editing capabilities. Experiments show that TailSteer can greatly reduce drift on FLUX, SenseNova, and Qwen series models in 15-20 rounds of editing and largely maintains the performance of the base model on large text-to-image and single-image editing benchmarks.
NVIDIA Technical BlogBlack Forest Labs's FLUX 3 Image Klein accelerates image generation on NVIDIA Blackwell GPUFLUX 3 Image Klein achieves a 3.5x speedup on NVIDIA Blackwell GPU. The model parameter size is 175B, scoring 95.3 on MMLU evaluation, supporting 8k context length, and providing closed-source services via API.
