Tag
long context
Long context refers to an LLM’s ability to keep and use very large histories in one pass, shaping memory design, retrieval, fast-weight updates, and stable reasoning. It shows up in 1M-2M token windows, state-space memory, TTT, and agent workflows.
19 articles
CoinRAG Reuses Fine-Grained KV Caches for RAG
CoinRAG cuts long-context RAG prefill cost by reusing fine-grained nugget KV caches instead of full chunks.
SALA Boosts Long Context on Edge AI
A mixed attention design lets edge models handle longer context with lower compute cost.
ReToken uses one learned token to fetch visual context
ReToken uses one learned token to retrieve relevant visual tokens and improve long-context vision-language retrieval.
Kimi K3 Proves Intelligence Still Costs Too Much
Kimi K3 is a top-tier intelligence model, but its price keeps it from being the obvious buy.
RoboTTT scales robot memory to 8K timesteps
RoboTTT pushes robot policy context to 8K timesteps and improves long-horizon control without raising inference latency.
Kimi’s long-context push keeps getting bigger
Moonshot AI’s Kimi chatbot keeps expanding context, agents, and model size, with Kimi K2.5 arriving in January 2026.
Kimi K2.6 turns agents into a swarm
Kimi K2.6 is an open-source multimodal agent model for long coding runs, UI generation, and swarm-style task orchestration.
Self-host MiniMax M3 on GPU cloud
MiniMax M3 brings 229.9B MoE weights, 1M context, and multimodal output, but it needs serious GPU memory to run.
Gemma 4 brings 256K context to open models
Google’s Gemma 4 adds text, image, and audio input, plus up to 256K context and five model sizes for local or server use.
MiniMax M3 adds 1M-token coding power
MiniMax M3 brings coding and agent features, a 1 million-token context window, and multimodal input to the company’s flagship model.
2026 LLM paper lists are a better research tool than feeds
Curated LLM paper lists beat raw feeds because they turn scattered research into usable context.
Best Kimi Models in 2026: K2.5 vs K2 Thinking
Kimi K2.5 leads Moonshot AI’s 2026 lineup with 256K context, 1T parameters, Agent Swarm Mode, and low API pricing.
Gemini turns Google’s AI stack into one app
A developer’s breakdown of Gemini’s rollout, model tiers, and why Google folded search, app, and Vertex AI into one AI surface.
MiniMax-M1 brings 1M-token open reasoning model
MiniMax released M1, an open-source reasoning model with 1M-token context, 80k output, and low-cost API pricing.
Sessa: Attention and State-Space Memory for Long Context
Sessa mixes attention with recurrent state-space feedback to improve long-context recall, with power-law memory tails and strong benchmark results.
In-Place TTT Lets LLMs Adapt at Inference
A new test-time training setup lets LLMs update fast weights in place, aiming for better long-context adaptation without full retraining.
Grok 4.20: xAI's new flagship model explained
xAI’s Grok 4.20 adds a 2M-token context window, multi-agent reasoning, and API pricing from $2 per million input tokens.
Gemini 3.1 Pro: Google’s new top model in numbers
Gemini 3.1 Pro posts 77.1% on ARC-AGI-2, 94.3% on GPQA Diamond, and a 1M-token context window, while keeping Gemini 3 pricing.
Universal YOCO aims to scale depth without cache bloat
YOCO-U mixes recursive computation with efficient attention to scale LLM depth while keeping inference overhead and KV cache growth in check.