AI Briefing#long-context
The latest AI and developer news about #long-context, with the original source and a short summary.
The latest AI and developer news about #long-context, with the original source and a short summary.
Running Qwen3.6-27B on a single RTX 3090 achieved 85 TPS at 125K context.
Qwen3.5-Omni achieves SOTA with 256k context and over 100 million hours of audio-visual data.
Reproduction code and benchmarks for Cartridges and STILL on a single GPU have been released.
Open-source reproduction code for Cartridges and STILL, KV-cache compression techniques for long-context inference, has been released.
Compressed the KV cache on Qwen 3.6's 1M context, reducing it from 10.7GB to 6.9GB.
SwiftLM presented results of running a 35B Qwen model on a 24GB Mac with 40K~100K context.
ClaudeDevs fixed an Opus 4.7 long-context bug and reset weekly and 5-hour limits.
Opus 4.7 improved on coding but regressed significantly on long-context reasoning.
In MRCR v2 8-needle at 1M tokens, Opus 4.7 scores 32.2%, significantly lower than 4.6's 78.3%.
Claude Opus 4.7 has been added to Bedrock, strengthening coding, agent, and document work.
A look at the internal culture and growth story of Moonshot AI (Kimi), which is rapidly growing as it surpasses a $16 billion valuation.
AI21 hosted a hackathon leveraging Jamba's 256K context window, showcasing innovative AI use cases.
AI21 Labs announced the Jamba 1.5 models, which maximize effective context length and improve cost efficiency.
Long context technology is drawing attention as a key factor determining the scalability and accuracy of enterprise AI.
Hugging Face has integrated Ulysses Sequence Parallelism, which supports long-context training of over a million tokens, into its major libraries.
AI21 has proposed Boring Agents, which prioritize predictability and auditability.
Anthropic has revealed two prompting techniques that maximize information recall performance in Claude's long context environments.
Jamba Reasoning 3B was much faster than Qwen3 4B on a 60K-token QA task.
AI21 has released Jamba Reasoning 3B, a 3B-class on-device open-source reasoning model.
TextQuests, a text-based game benchmark for evaluating LLMs' reasoning and exploration abilities as agents, has been announced.
Since Mamba, hybrid LLMs have been replacing the limits of Transformers.
A planning-based AI pipeline was presented as a way to manage both product description quality and regulation together.
It presents an AI system that cross-checks regulatory changes against contracts and policies in real time to find compliance gaps.