https://tpsreport.news/news/scaledown-summarize-1m-context-openrouter TPS — Tokens Per Second en 2026-10-13T12:20:51.387Z ScaleDown launches Summarize, a 1M-context abstractive summarization model at $0.04 per 1M input tokens on OpenRouter ScaleDown, Summarize, summarization, OpenRouter, RAG, long context, model release, document processing https://tpsreport.news/news/minicpm-v-4-7-1b-on-device-multimodal-openbmb TPS — Tokens Per Second en 2026-10-13T11:21:18.065Z OpenBMB releases MiniCPM-V 4.7 1B, an on-device vision model with 4x/16x visual token compression MiniCPM-V, OpenBMB, multimodal, on-device AI, small language models, vision-language model, open weights, Hugging Face https://tpsreport.news/news/anthropic-claude-reasoning-token-distillation-defenses TPS — Tokens Per Second en 2026-10-13T11:20:45.291Z Anthropic Says New Defenses Cut Claude Reasoning-Token Extraction by 90–99% Anthropic, Claude, distillation, reasoning tokens, API security, Zhipu, Moonshot, GLM-5.3 https://tpsreport.news/news/anthropic-study-glm-5-2-ai-worm-replication-consumer-devices TPS — Tokens Per Second en 2026-10-13T10:50:49.398Z Anthropic study: GLM-5.2 completes full AI self-replication chain on $55 TV boxes in 4 of 5 runs Anthropic, AI safety, GLM-5.2, AI worm, self-replication, IoT security, Gemma, Qwen3.5 https://tpsreport.news/news/gemini-spark-task-redesign-main-chat-app TPS — Tokens Per Second en 2026-10-13T04:50:41.601Z Google folds Gemini Spark agent into main Gemini app as 'Task' button next to Chat Gemini, Gemini Spark, Google, AI agents, Gemini app, product update, Android https://tpsreport.news/news/minicpm-v-4-7-35b-a3b-on-device-moe TPS — Tokens Per Second en 2026-10-13T03:51:02.847Z OpenBMB's MiniCPM-V 4.7 35B-A3B MoE claims 13+ tokens/s on iPhone 17 Pro Max in 3 GB MiniCPM-V, OpenBMB, mixture-of-experts, on-device AI, multimodal, open-weights, vision-language, Hugging Face https://tpsreport.news/news/sakana-ai-mass-self-improving-qwen3-6-27b-blind-spots TPS — Tokens Per Second en 2026-10-12T18:06:00.521Z Sakana AI's MASS self-improvement lifts Qwen3.6-27B win rate to 69.9%, but self-judging passes flawed runs Sakana AI, UC Berkeley, MASS, recursive self-improvement, Qwen3.6-27B, multi-agent, self-supervision, research https://tpsreport.news/news/meta-muse-wonder-agent-model-paid-subscription TPS — Tokens Per Second en 2026-10-12T17:20:49.008Z Meta launches Muse Wonder agent model, gating it behind $20 and $100 monthly Muse app plans Meta, Muse Wonder, Muse, AI agents, Muse Code, Meta Model API, Alexandr Wang, subscriptions https://tpsreport.news/news/karpathy-four-tiers-ai-users-5000-internal-frontier-access TPS — Tokens Per Second en 2026-10-12T13:05:49.284Z Karpathy estimates only 5,000 people worldwide have access to internal frontier AI systems Andrej Karpathy, Anthropic, OpenAI, AI agents, frontier models, AI adoption, Sam Altman, reinforcement learning https://tpsreport.news/news/tasteval-opus-5-5-research-taste-benchmark TPS — Tokens Per Second en 2026-10-12T12:50:55.705Z TasteVal: Opus 5.5 beats expert baseline at AI research experiment design, with a 2.30x compute multiplier benchmark, TasteVal, Humanity's Sixth Sense, Opus 5.5, GPT 6 Astra, recursive self-improvement, Kimi K3, GLM-5.3-flash https://tpsreport.news/news/alphaproof-nexus-erdos-problems-science-peer-review TPS — Tokens Per Second en 2026-10-12T09:50:46.953Z DeepMind's AlphaProof Nexus solves 9 of 353 open Erdős problems in peer-reviewed Science paper AlphaProof Nexus, Google DeepMind, Gemini 3.1 Pro, Lean, Erdős problems, mathematics, formal verification, Science https://tpsreport.news/news/agent-teams-vs-solo-agents-vibe-code-bench-vals-ai TPS — Tokens Per Second en 2026-10-11T15:50:46.963Z Vals AI: Agent teams cost 1.8x–5.1x more than solo agents, with one significant gain in four tests agent teams, multi-agent, Vals AI, Vibe Code Bench, GPT-6 Sol, Claude Opus 5.5, token cost, benchmark