Thursday, Nov 5

8 items · Updated 11:31 AM ET

Showing the Nov 5 brief. Back to today ›

2 new rumors today →
Top stories
2OfficialProduct

llama.cpp v0.9.0 adds OCP FP8 E4M3 quant, multi-sequence KV flash-attn, and four new architectures

Ggml-org published this on Nov 4, 2026 (source: github.com).

ggml-org released llama.cpp v0.9.0 with OCP FP8 E4M3 quantization, a new flash_attn_ext_rows op aimed at efficient unified KV cache multi-sequence inference, and a generic batched tensor copy API. The notes add four architectures (Clover, Limite, Metis-1.6, and Mistral Large 4), expand audio generation and transcription with Chatterbox TTS, Nemotron 3.5 Transcribe, and predefined Pocket TTS voices, and cite major Gemma 4 E2B/E4B prefill optimizations alongside a ggml bump.

Part of llama.cpp
Source: ggml-org
3OfficialProduct

Together AI launches Together Compiler for continual agentic optimization of AI inference

Together AI published this on Nov 5, 2026 (source: together.ai).

Together AI introduced Together Compiler (TKCC), described as a continual optimization loop that uses high-throughput agents and persistent engines to tune inference across shifting models, hardware, and traffic shapes. The company says that in under 24 hours TKCC reached up to 2.6× higher tokens per second on frontier open models (including Kimi K3, GLM 5.3, MiniMax M3.1-Flash, and DeepSeek v4.1-Flash), posted SemiAnalysis InferenceX low-latency marks of 1111 TPS on MiniMax M3.1-Flash and 1205 TPS on DeepSeek v4.1-Flash, and beat available open-source megakernels by about 1.7× on a Llama 1B batch-size-1 setting, enabled by roughly 252× higher experiment throughput versus baseline agentic optimizers. Together says the compiler is already serving production traffic and also generalizes beyond LLM inference.

Part of Together Compiler
Source: Together AI
4OfficialProduct

vLLM adds day-0 support for Moonshot’s Kimi K3.1 Pro sparse-attention MoE

VLLM Blog published this on Nov 4, 2026 (source: vllm.ai).

The vLLM team and Inferact published day-0 serving support for Moonshot AI’s Kimi K3.1 Pro, a multimodal MoE cited at about 2.8T parameters (104B active) with a 1M-token context and Sparse Attention (DSA) on its 24 full-attention layers. The write-up covers a packed KV cache for MLA latents, indexer keys, and KDA state with FP8 or NVFP4 latents and optional MXFP4 index keys, token-sharded prefill, decode context parallelism with a sparse indexer, and a fused low-latency DSA decode path, with support paths called out for NVIDIA, AMD, Google TPUs, and Alibaba T-Head.

Source: vLLM Blog
Latest
OfficialSalesforce says Agentforce Coworker usage grew 21× in three months to more than 185,000 users

Salesforce published this on Nov 4, 2026 (source: salesforce.com).

Salesforce announced a major expansion of Agentforce Coworker, saying more than 185,000 users are putting the AI teammate to work after about 21× usage growth in three months, with customers such as The Adecco Group, Fulton Bank, and BACA Systems expanding adoption. New capabilities include reusable Skills, conversation history, upcoming scheduled tasks, Microsoft Teams/mobile/desktop surfaces, and predictable pricing at $50 per user per month, with Coworker drawing Salesforce records and Data 360 context to coordinate Agentforce agents.

Part of Salesforce
Source: Salesforce
OfficialAWS says U Mobile runs agentic telco operations on Amazon Bedrock AgentCore

AWS published this on Nov 4, 2026 (source: aws.amazon.com).

AWS published a customer story on Malaysia’s U Mobile building AiDaH, an agentic AI operations platform on Amazon Bedrock AgentCore for autonomous action across telco operations. Nine months after production launch, AWS says AiDaH handles more than 150 operational scenarios across six divisions and cut trouble tickets by up to 84.5%.

Part of Bedrock AgentCore
Source: AWS
OfficialGoogle publishes a health-AI evaluation framework covering accuracy, safety, and usefulness

Google Health Blog published this on Nov 4, 2026 (source: blog.google).

Google’s health blog, with health experts Dr. Elizabeth McGlynn and Dr. Dana Gelb Safran, published a whitepaper on an end-to-end evaluation approach for AI use cases that address people’s health needs. The framework assesses models across accuracy, safety, and usefulness; Google says it uses the methodology on its own models and is sharing it to promote transparency for the broader health ecosystem.

Source: Google Health Blog
How we label
Official
An institution announces something that has already happened, through its own channel or a founder or executive account.
Statement by …
A named person's views, predictions, plans or comments, including remarks reported by media when the original is unavailable.
Reported by …
Only media or another third party is saying it. We name the outlet.
Unconfirmed
Circulating online (posts, comments, leaks). No company, named person or outlet has confirmed it.

Items older than 72 hours never go in Top stories and show their original date.

DailyCitedRSSAboutPrivacy