https://tpsreport.news/news/neurodx-sphota-240m-model-liquid-ai-benchmark TPS — Tokens Per Second en 2026-11-02T07:50:53.845Z NeuroDX's 240M-parameter Sphota scores 52.4% vs Liquid AI's 52.1% across about 11 benchmarks, company claims Sphota, NeuroDX, Manas AI, Liquid AI, IndiaAI Mission, small language models, benchmarks https://tpsreport.news/news/semianalysis-gdpx-benchmark-opus-5-5-12-percent TPS — Tokens Per Second en 2026-11-02T05:35:49.332Z SemiAnalysis launches GDPx agentic benchmark across 11 domains; best model, Opus 5.5, scores 12% GDPx, SemiAnalysis, benchmark, Opus 5.5, GPT-6 Astra, Gemini 4 Carbon, agentic, GDPval https://tpsreport.news/news/gpt-6-2-astra-gray-test-claude-mythos-5-5-spotted TPS — Tokens Per Second en 2026-11-02T05:20:55.122Z GPT-6.2-Astra enters gray test for some Pro users as Claude Mythos 5.5 appears in site code GPT-6.2-Astra, Claude Mythos 5.5, OpenAI, Anthropic, gray test, Hugging Face, leak https://tpsreport.news/news/deepseek-dsh-superagent-open-source-coding-agent-prototype TPS — Tokens Per Second en 2026-11-02T04:51:02.329Z DeepSeek's Harness team open-sources dsh-superagent, a research prototype coding agent driven by GitHub DeepSeek, DeepSeek Harness, dsh-superagent, coding agent, open source, GitHub, Feishu https://tpsreport.news/news/meituan-longcat-2-5-1-6t-multimodal-agentic-coding TPS — Tokens Per Second en 2026-11-02T02:35:57.281Z Meituan releases LongCat-2.5: 1.6T-parameter multimodal model, 88.4 on Terminal Bench v2.1 Meituan, LongCat-2.5, agentic coding, multimodal, Terminal Bench, W4A4, mixture of experts https://tpsreport.news/news/liquid-ai-d1-2b-decision-model TPS — Tokens Per Second en 2026-11-01T17:06:23.904Z Liquid AI releases d1-2B, a 1.92B decision model that answers in one forward pass with no generated tokens Liquid AI, d1-2B, LFM3, decision model, open weights, vision-language, classification https://tpsreport.news/news/adobe-editagent-image-editing-agent-verification TPS — Tokens Per Second en 2026-11-01T15:05:44.798Z EditAgent: Adobe, UC San Diego agent beats trained edit judges with measured image data Adobe, EditAgent, image editing, AI agents, UC San Diego, Qwen3.8, EditScore, research https://tpsreport.news/news/claude-opus-5-5-system-prompt-game-cloning-rule TPS — Tokens Per Second en 2026-11-01T13:35:57.706Z Anthropic adds two-sentence rule to Opus 5.5 system prompt to stop Claude cloning existing games Anthropic, Claude Code, Claude Opus 5.5, system prompt, game cloning, copyright https://tpsreport.news/news/vals-ai-pro-china-lean-chinese-models-study TPS — Tokens Per Second en 2026-11-01T12:50:44.616Z Vals AI study: Chinese models' pro-China lean is lower than expected, and shifts with prompt language Vals AI, benchmark, Chinese AI models, Kimi K3, GLM-5.3 Flash, Qwen3.8 Max, DeepSeek, political bias https://tpsreport.news/news/z-ai-tang-jie-self-improving-ai-agi-safety TPS — Tokens Per Second en 2026-11-01T12:20:46.465Z Z.ai founder Tang Jie says self-improving AI is a key step toward AGI, but must be done safely Z.ai, Tang Jie, AGI, recursive self-improvement, AI safety, China AI https://tpsreport.news/news/streamvibe-google-proactive-video-agent-benchmark TPS — Tokens Per Second en 2026-11-01T11:35:45.282Z Google-led StreamVibe video agent decides when to speak up, but all tested models degrade over long videos Google, StreamVibe, video agents, proactive assistants, Qwen3.5, Gemini, benchmark, research https://tpsreport.news/news/chatgpt-chat-work-merger-shared-token-budget-december TPS — Tokens Per Second en 2026-11-01T10:05:44.921Z ChatGPT to share one token budget across Chat and Codex on Plus, Pro and Business from December OpenAI, ChatGPT, Codex, usage limits, pricing, ChatGPT Plus, ChatGPT Pro, ChatGPT Business https://tpsreport.news/news/nace-ai-drex-max-v1-27b-apple-silicon TPS — Tokens Per Second en 2026-11-01T08:20:48.918Z Nace AI open-sources Drex Max v1, a 27B multimodal decision model for Apple Silicon Drex Max v1, Nace AI, open source, Apple Silicon, local inference, multimodal, 27B https://tpsreport.news/news/sensenova-loop-9b-preview-looped-transformer-apache-2 TPS — Tokens Per Second en 2026-11-01T02:05:58.024Z SenseTime open-sources 9B SenseNova-Loop-9B-Preview, a looped model, under Apache 2.0 SenseTime, SenseNova-Loop-9B-Preview, open-source, looped transformer, multimodal, Apache 2.0, Hugging Face https://tpsreport.news/news/anthropic-claude-watermark-opus-4-6-sonnet-4-6-sampling-loophole TPS — Tokens Per Second en 2026-10-31T19:35:44.489Z Anthropic watermarks four more Claude models, but sampling tweaks defeat it on Opus 4.6 and Sonnet 4.6 Anthropic, Claude, watermark, SynthID, EU AI Act, Claude Opus 4.6, Claude Sonnet 4.6, Claude Mythos 5 https://tpsreport.news/news/deepseek-super-agent-router-worker-architecture TPS — Tokens Per Second en 2026-10-31T19:05:52.059Z DeepSeek open-sources Super-agent prototype: Router-Worker design hits 15.7-minute P95 in simulation DeepSeek, agents, Super-agent, open-source, agent-architecture, GitHub, DeepSeek Harness https://tpsreport.news/news/anthropic-claude-generated-datasets-train-models-sabotage-oversight TPS — Tokens Per Second en 2026-10-31T15:20:50.679Z Anthropic's Claude wrote harmless-looking datasets that trained models to sabotage oversight Anthropic, Claude, AI safety, alignment, red teaming, emergent misalignment, fine-tuning, data monitoring https://tpsreport.news/news/anthropic-self-replicating-ai-botnets-experts-respond TPS — Tokens Per Second en 2026-10-31T13:05:51.266Z Anthropic Warns Small AI Models Are Months From Self-Replicating Botnets; Experts Call Threat Familiar Anthropic, AI safety, botnets, self-replication, cybersecurity, Mirai, Frontier Red Team, open-weight models https://tpsreport.news/news/claude-opus-5-5-game-recreation-reverse-engineering-refusals TPS — Tokens Per Second en 2026-10-31T12:35:46.240Z Users report Claude refusing game recreation and reverse-engineering tasks; Anthropic silent Anthropic, Claude Code, Claude Opus 5.5, reverse engineering, Cyber Verification Program, Project Glasswing, safety policy, game development https://tpsreport.news/news/factory-droid-1-0-model-independent-coding-agent-harness TPS — Tokens Per Second en 2026-10-31T12:20:46.140Z Factory releases Droid 1.0 agent harness, 65.2% on Terminal-Bench 4 with Claude Opus 5.5 Factory, Droid, coding agent, Terminal-Bench 4, Claude Opus 5.5, agent harness