https://tpsreport.news/news/prismml-ternary-bonsai-2-27b TPS — Tokens Per Second en 2026-09-18T19:35:58.943Z PrismML Releases Ternary Bonsai 2 27B, a Compressed Reasoning Model with 262K Context PrismML, Ternary Bonsai 2, reasoning model, model compression, Qwen3.8, OpenRouter https://tpsreport.news/news/claude-code-agents-md-support-mods TPS — Tokens Per Second en 2026-09-18T19:20:45.628Z Claude Code 2.1.277 Adds AGENTS.md Support Via New Mods System claude-code, agents-md, anthropic, coding-agents, developer-tools https://tpsreport.news/news/ios-27-apple-intelligence-20-features TPS — Tokens Per Second en 2026-09-18T19:06:28.491Z Apple Ships iOS 27 with Rebuilt Siri and 20+ New Apple Intelligence Features Apple Intelligence, iOS 27, Siri, Apple, product update https://tpsreport.news/news/typesafe-ai-jev-probability-model TPS — Tokens Per Second en 2026-09-18T19:06:02.549Z OpenAI RLHF Co-Inventor Launches Jev, a Non-LLM Model That Outputs Probabilities Instead of Text TypeSafe AI, Jev, non-LLM model, AI classification, model routing, RLHF, Diogo Almeida https://tpsreport.news/news/google-cc-ai-agent-family-household-management TPS — Tokens Per Second en 2026-09-18T17:50:52.749Z Google Relaunches CC as a Family-Management AI Agent Running on Gemini and Antigravity Google, AI agents, Gemini, household management, consumer AI, productivity https://tpsreport.news/news/glm-5-3-flashx-200-tokens-per-second TPS — Tokens Per Second en 2026-09-18T17:06:18.814Z Z.ai Releases GLM-5.3-FlashX, a 200 Tokens/Second Variant of Its GLM-5.3-Flash Model Z.ai, GLM-5.3, GLM-5.3-FlashX, Zhipu AI, open source models, multimodal, long context https://tpsreport.news/news/kimi-k3-amazon-bedrock-2-8-trillion-parameters TPS — Tokens Per Second en 2026-09-18T17:05:54.283Z Moonshot AI's 2.8 Trillion-Parameter Kimi K3 Launches on Amazon Bedrock with 1M-Token Context Kimi K3, Moonshot AI, Amazon Bedrock, AWS, open-weight models, prompt caching, large language models https://tpsreport.news/news/aws-bedrock-agentcore-runtime-update-cold-starts TPS — Tokens Per Second en 2026-09-18T15:50:47.258Z AWS Overhauls Bedrock AgentCore Runtime, Cuts Cold Starts to Flat 2 Seconds Regardless of Image Size AWS, Amazon Bedrock, AgentCore, AI agents, infrastructure, serverless https://tpsreport.news/news/meta-muse-ai-agent-mac-desktop-app TPS — Tokens Per Second en 2026-09-18T15:35:50.348Z Meta Brings Muse AI Agent to Mac, Enabling Actions Across Files, Mail and Calendar Meta, Muse, AI agents, desktop app, consumer AI, Mac https://tpsreport.news/news/deepmind-chain-of-thought-transparency-safety-risk TPS — Tokens Per Second en 2026-09-18T14:50:56.604Z DeepMind Institute Warns AI Chain-of-Thought Transparency Is Eroding, Citing GPT-6 Astra Monitoring Drop AI safety, chain-of-thought, interpretability, DeepMind Institute, model alignment, GPT-6, Gemini 3 https://tpsreport.news/news/openai-astra-for-law-legal-research TPS — Tokens Per Second en 2026-09-18T13:50:52.412Z OpenAI Launches Astra for Law, a Legal Research Tool Built on GPT-6 Astra OpenAI, GPT-6 Astra, legal AI, product launch, legal tech, Harvey, Legora https://tpsreport.news/news/atria-dawn-preview-shanghai-ai-lab-agentic-model TPS — Tokens Per Second en 2026-09-18T10:20:59.962Z Shanghai AI Lab Releases Atria Dawn Preview, a 744B-Parameter MoE Agentic Model Built on GLM-5.2 Atria Dawn Preview, Shanghai AI Lab, internlm, GLM-5.2, agentic AI, MoE model, open weights https://tpsreport.news/news/moonshot-kimi-financial-services-data-partners TPS — Tokens Per Second en 2026-09-18T08:50:46.585Z Moonshot AI Launches Kimi for Financial Services With S&P, Wind, EDGAR Data Access Moonshot AI, Kimi, financial services, China AI, S&P Global, product launch https://tpsreport.news/news/safari-27-mcp-server-ai-agents TPS — Tokens Per Second en 2026-09-18T02:35:43.939Z Safari 27 Adds Native MCP Server, Letting AI Agents Like Claude Code and Codex Control the Browser Safari, WebKit, MCP, Apple, developer tools, Claude Code, Codex, browser automation https://tpsreport.news/news/xing4-29b-a4b-ascend-npu-release TPS — Tokens Per Second en 2026-09-18T01:50:52.211Z China Telecom Releases Xing4.0-29B-A4B, a 29B MoE Model Trained Entirely on Ascend NPUs China Telecom, Xing4.0, MoE, Ascend NPU, MindSpore, open source, TeleChat, agentic AI https://tpsreport.news/news/openai-python-sdk-v3-15-0 TPS — Tokens Per Second en 2026-09-18T01:05:42.760Z OpenAI Python SDK v3.15.0 Adds Managed WebSocket Sessions and Prompt-Cache Prewarming OpenAI, Python SDK, API, changelog, WebSocket, prompt caching, agents https://tpsreport.news/news/unbiased-pareto-multimodal-model TPS — Tokens Per Second en 2026-09-17T23:20:48.372Z Unbiased Launches Pareto, a $2.50/$7.50-per-Million-Token Multimodal Model for Coding and Agents Pareto, Unbiased, multimodal model, OpenRouter, model release, coding agents https://tpsreport.news/news/prismml-bonsai-2-27b-compressed-llm TPS — Tokens Per Second en 2026-09-17T22:50:58.452Z PrismML's Bonsai 2 Compresses 27B-Parameter Model to 5.9GB, Retains 98% of Benchmark Performance PrismML, model compression, LLM, on-device AI, Qwen, ternary quantization, Bonsai 2 https://tpsreport.news/news/openai-self-injected-prompt-compaction-summary TPS — Tokens Per Second en 2026-09-17T21:05:51.964Z OpenAI Discloses Case of Model Injecting Fake Jailbreak Persona Into Its Own Context Summary openai, prompt-injection, ai-safety, model-alignment, agents, reinforcement-learning, compaction https://tpsreport.news/news/openai-models-hid-notes-for-successors TPS — Tokens Per Second en 2026-09-17T20:50:54.587Z OpenAI Discloses Its Models Secretly Coached Future Versions to Hide Mistakes OpenAI, AI alignment, AI safety, GPT-5.6, misalignment, model deception https://tpsreport.news/news/claude-code-projects-multi-agent-cloud TPS — Tokens Per Second en 2026-09-17T19:05:44.473Z Anthropic Relaunches Claude Code Projects to Coordinate Multiple Cloud Agents Claude Code, Anthropic, AI agents, multi-agent systems, coding tools, product update https://tpsreport.news/news/claude-code-parallel-agent-threads-update TPS — Tokens Per Second en 2026-09-17T18:50:45.276Z Anthropic Rebuilds Claude Code Projects to Run Parallel AI Agent Threads Anthropic, Claude Code, AI agents, coding automation, developer tools https://tpsreport.news/news/siri-ai-ios-27-125-tests TPS — Tokens Per Second en 2026-09-17T18:06:06.762Z iOS 27's Rebuilt Siri AI Handles 125 Real-World Tasks in Independent Test Siri, Apple, iOS 27, Siri AI, voice assistant, David Pogue https://tpsreport.news/news/wood-mackenzie-apex-amazon-bedrock-agentcore TPS — Tokens Per Second en 2026-09-17T15:50:56.547Z Wood Mackenzie Builds Shared Agentic Platform APEX on Amazon Bedrock AgentCore Amazon Bedrock, AgentCore, Wood Mackenzie, AWS, agentic AI, enterprise AI, AI infrastructure https://tpsreport.news/news/gpt-6-astra-gaming-benchmarks-arc-agi-3 TPS — Tokens Per Second en 2026-09-17T14:51:46.292Z OpenAI's GPT-6 Astra Beats Pokémon in 18 Hours, Scores 62.7% on ARC-AGI-3 OpenAI, GPT-6, AI agents, ARC-AGI-3, benchmarks, reasoning models https://tpsreport.news/news/instinct-meta-muse-ai-agents-calling-feature TPS — Tokens Per Second en 2026-09-17T14:05:47.898Z Instinct and Meta's Muse AI Agents Both Add Phone-Calling Features Within Hours of Each Other AI agents, Instinct, Meta Muse, voice AI, product launch, consumer AI https://tpsreport.news/news/openai-misalignment-framework-prompt-injection-notes TPS — Tokens Per Second en 2026-09-17T13:51:17.184Z OpenAI Launches Framework to Disclose AI Misalignment, Reveals Model Injected Fake Instructions Into Its Own Notes OpenAI, AI safety, misalignment, prompt injection, alignment research, GPT-6, AI transparency https://tpsreport.news/news/openai-misalignment-reports-model-behavior TPS — Tokens Per Second en 2026-09-17T11:05:52.049Z OpenAI Discloses Six Cases of Models Faking Data, Hiding Behavior From Testers OpenAI, AI safety, alignment, misalignment reports, GPT-5.6 Sol, AI transparency https://tpsreport.news/news/gpt-6-astra-enigma-decryption-claim TPS — Tokens Per Second en 2026-09-17T10:05:59.897Z Bloomberg Developer Says OpenAI's GPT-6 Astra Cracked an 83-Year-Old Nazi Enigma Message in 10 Hours GPT-6 Astra, OpenAI, Enigma, cryptography, AI agents, historical decryption https://tpsreport.news/news/github-copilot-runtime-rust-migration-1 TPS — Tokens Per Second en 2026-09-17T00:36:15.077Z GitHub Used Copilot to Rewrite Its Own Agent Runtime in 800,000 Lines of Rust GitHub Copilot, Rust, agentic coding, software engineering, GitHub https://tpsreport.news/news/github-copilot-runtime-rust-migration TPS — Tokens Per Second en 2026-09-17T00:35:48.255Z GitHub Used Copilot to Rewrite Its Own Copilot Runtime in 800,000 Lines of Rust GitHub Copilot, Rust, agentic coding, software migration, developer tools