Long Context refers to models and techniques that process very large inputs — many thousands to millions of tokens — in a single pass, letting a system reason over whole documents, codebases, or conversations at once.
A large window is not free capability: attention can weaken toward the middle of long inputs (the lost-in-the-middle effect), latency and cost rise with length, and more tokens do not guarantee better answers. Bigger is not automatically a substitute for retrieval.
For builders the practical question is long context versus RAG: loading everything is simpler but slower and pricier, while retrieval is leaner but can miss. Most production systems combine them, using Context Engineering to place only the most relevant material into even a large window.
Topic: LLMs Related: Context Engineering RAG
Recent Updates
- 2026-08-12: Cracks in the Foundation: Seemingly Minor Architectural Choices Impact Long Context Extension (cs.CL updates on arXiv.org) · arxiv.org — Olmo · LLaMA · Qwen · LLaMA-3
- 2026-08-12: Four Architecture Decisions Reduce Model Long-Context Performance by 47 Percent (DAIR.AI) · arxiv.org — Ai2 · Carnegie Mellon University University of Washington · LLaMA · Qwen · Olmo
- 2026-08-13: How Compaction Works in Pi (Hacker News) · earendil.com
- 2026-08-15: Qwen3.8-27B Open-Sourced: Outperforms Claude Opus on Coding and Agents, Deployable on Consumer GPUs (量子位) · qbitai.com — Alibaba · Anthropic · Hugging Face · NVIDIA 梦瑶 · Qwen3.8-27B Claude Opus 4.6 Max
- 2026-08-17: Nanbeige4.2-3B on Apple Silicon: Fixing Deployment Bugs and Decreasing Looped Transformer Memory Overhead (cs.AI updates on arXiv.org) · arxiv.org — Hugging Face · Apple · Nanbeige4.2-3B
- 2026-08-17: Qwen3.8 27B Local Inference: System-Level Optimization Yields 50 tok/s at 256K Context (Hacker News) · piszczek.pl — NVIDIA · Qwen3.8-27B
- 2026-08-18: Muse Glimmer is a memory hierarchy disguised as a 30B Transformer (Hacker News) · abstractextraordinary.com — Meta · Muse Glimmer
- 2026-08-18: Maglev: Sliding Recurrent Memory (alphaXiv) · alphaxiv.org
- 2026-08-24: SDAD: Spec-Driven Agentic Development for the AI-Native SDLC (breakingnewsofficial) · arxiv.org
- 2026-08-24: Inhibitory Attention for Clinical Long-Context Reasoning: Characterizing and Mitigating Lost-in-the-Middle Effects in EHR Processing (breakingnewsofficial) · arxiv.org — Qwen2.5-7B-Instruct
- 2026-08-26: GLM-5.3 Flash: Hybrid Attention Architecture Delivers Frontier Performance at 1/40 the Cost (breakingnewsofficial) · qbitai.com — Zhipu · OpenRouter · OpenCode Tim Jayas · GLM-5.3-Flash · GLM-5.2 · GLM-4.5 · Claude Opus 4.8
- 2026-08-27: Understanding the Energy Scaling of Large Language Models Inference Across Context Lengths and Attention Architectures (breakingnewsofficial) · arxiv.org — NVIDIA
- 2026-08-29: Why Local LLM Deployments Produce Different Results: Inference Stack Variations and Numerical Drift (breakingnewsofficial) · qbitai.com — NVIDIA · Hugging Face thr3e · Qwen3.6-27B · Qwen3.8
- 2026-09-01: Claude Fable 5.1 Released — Agentic AI Model with 1M Context and Reduced Cache Costs (breakingnewsofficial) · platform.claude.com — Anthropic · Amazon · Google · Microsoft · AWS Anton Vdovitchenko · Claude Fable 5.1 Claude Mythos 5.1 · Claude Fable 5 · Claude Opus 5 · Claude Opus 4.8
- 2026-09-03: NeoMME: Efficient Multilingual Multimodal Encoder from Hugging Face (breakingnewsofficial) · huggingface.co — Hugging Face · NVIDIA NeoMME · ColPali SigLIP2 · ModernBERT ModernVBERT
- 2026-09-04: Reasoning Traces as State for Better Long-Context Encoding (breakingnewsofficial) · Z.AI — DeepSeek V4 Pro
- 2026-09-05: GPT-6 Astra Shows Strongest Gains in Cross-File Code Review Despite Higher Costs (breakingnewsofficial) · coderabbit.ai — OpenAI · CodeRabbit · Anthropic · GPT-6 Astra · GPT-5.6 Sol · Opus 5 · Fable 5.1 · Terra · Luna
- 2026-09-06: Embodied AI Startups Adopt In-Context Learning as New Scaling Path (breakingnewsofficial) · qbitai.com — Skild AI Generalist AI COCO Matrix Fourier Intelligence Gao Yuxiang Li Fei-Fei Jim Fan Yuke Zhu S1 GEN-1.5 RoboTTT GPT-3
- 2026-09-07: KVMem: Virtualizing Million-Token Agent Workspaces on a Consumer GPU (breakingnewsofficial) · arxiv.org — Qwen3.8-27B · Qwen3.6-27B
- 2026-09-09: Memory in Deep Time-Series Models: A Unified Framework for Retention and Access (breakingnewsofficial) · arxiv.org
- 2026-09-11: Larger Context Window, Fewer Overcorrections: Optimizing Prompts and Batching for Minimal-Edit Grammatical Error Correction (breakingnewsofficial) · arxiv.org — Google Staruch · Gemini-3.1-Pro
FAQ
What is Long Context?
Long Context refers to model and system designs that process large input windows while preserving useful attention and answer quality. GROUNDING tracks context limits, compression, memory, and retrieval-vs-context tradeoffs.
Which topic does Long Context belong to?
On the GROUNDING radar, Long Context is grouped under the LLMs topic.
Which concepts are related to Long Context?
Related concepts tracked by the radar include Context Engineering, RAG.