https://tpsreport.news/news/iflytek-sparkaudio-1-0-99-languages-30b-a3b-moe TPS — Tokens Per Second en 2026-10-26T07:06:34.578Z iFLYTEK SparkAudio 1.0 speech model supports 99 languages and 202 dialects, pairs 0.65B encoder with 30B-A3B MoE iFLYTEK, SparkAudio, speech recognition, mixture-of-experts, robotics, embodied AI, Huawei Ascend, multilingual https://tpsreport.news/news/primotion-paimon-0-embodied-model-300ms-action-chunk TPS — Tokens Per Second en 2026-10-26T07:05:49.715Z Primotion's Paimon-0 cuts action-chunk inference from about 3,000 ms to 300 ms, company says Paimon-0, Primotion, embodied AI, robotics, world action model, RoboColiseum, FPGA, Shenzhen https://tpsreport.news/news/cline-desktop-v0-0-49-changes-pane-terminal-local-models TPS — Tokens Per Second en 2026-10-26T04:50:52.147Z Cline Desktop v0.0.49 adds changed-files pane, project file tree and integrated terminal Cline, Cline Desktop, developer tools, coding agents, ClinePass, Kimi K3.1 Pro, Ollama, LM Studio https://tpsreport.news/news/kimi-k3-1-pro-rate-limit-apology-25-percent-quota-reset TPS — Tokens Per Second en 2026-10-26T02:20:45.234Z Kimi resets 25% of monthly quota for all paid tiers after K3.1 Pro rate-limit and quota problems Kimi, Moonshot AI, K3.1 Pro, rate limits, quota, sub-agents, Kimi Code https://tpsreport.news/news/nanbeige4-3-33b-512k-context-agentic-teacher-model TPS — Tokens Per Second en 2026-10-25T15:21:36.185Z Nanbeige4.3-33B released: 33B open model with 512K context targets agentic coding and cowork tasks Nanbeige, Nanbeige4.3-33B, open-weights, agentic-coding, teacher-model, 512K-context, Hugging Face https://tpsreport.news/news/anthropic-oss-scanner-free-vulnerability-scanner TPS — Tokens Per Second en 2026-10-25T13:20:39.245Z Anthropic launches OSS Scanner, a free opt-in vulnerability scanner for critical open-source projects Anthropic, OSS Scanner, Claude Mythos, security, open source, vulnerability scanning https://tpsreport.news/news/openai-red-team-agents-cpu-cache-covert-channel-microvm TPS — Tokens Per Second en 2026-10-25T11:50:50.582Z OpenAI red-team agents built a CPU cache covert channel across microVM sandboxes with no network or shared files OpenAI, AI safety, red teaming, sandboxing, microVM, CPU cache, covert channel, GPT-6 Astra Cyber https://tpsreport.news/news/ai21-labs-lift-bench-scoped-constraints-instruction-following TPS — Tokens Per Second en 2026-10-25T10:50:44.331Z AI21 Labs releases LIFT-Bench, scoping each constraint to one document section AI21 Labs, LIFT-Bench, LIFT-Train, instruction following, benchmark, reinforcement learning, reward hacking https://tpsreport.news/news/jack-clark-self-improving-ai-2027-tasteval-benchmark TPS — Tokens Per Second en 2026-10-25T09:50:49.355Z Anthropic's Jack Clark moves self-improving AI forecast to 2027 on TasteVal results Anthropic, Jack Clark, TasteVal, P-Zero Research, recursive self-improvement, Opus 5.5, benchmark, AI research automation https://tpsreport.news/news/anthropic-single-agent-beats-subagents-migration-kit TPS — Tokens Per Second en 2026-10-25T09:20:48.086Z Anthropic recommends consolidating sub-agents into one agent that loads context on demand Anthropic, AI agents, sub-agents, Claude, agent architecture, skills, context engineering https://tpsreport.news/news/qwen-personal-agent-framework-model-memory-ecosystem TPS — Tokens Per Second en 2026-10-25T07:20:45.323Z Qwen outlines personal agent framework: 8,000+ atomic capability requirements, tiered memory, open service ecosystem Qwen, Alibaba, personal agent, Qwen Go!, agent memory, CNCC 2026 https://tpsreport.news/news/kimi-k3-1-pro-influencer-promotion-allegations TPS — Tokens Per Second en 2026-10-25T05:05:43.122Z Kimi K3.1 Pro debut draws scrutiny over allegations of undisclosed influencer promotion for Moonshot Kimi K3.1 Pro, Moonshot, Kimi, influencer marketing, benchmarks, VentureBeat https://tpsreport.news/news/cognition-devin-vulnerability-fixes-plugin-open-source TPS — Tokens Per Second en 2026-10-25T03:20:43.539Z Cognition open-sources Devin plugin with 3 MIT-licensed skills to patch PHP, Node.js, GitLab RCE flaws Cognition, Devin, cybersecurity, open-source, agent-skills, RCE, GitLab, Node.js https://tpsreport.news/news/gemini-4-argon-google-adk-optimization-11-hour-run TPS — Tokens Per Second en 2026-10-24T16:20:45.531Z Gemini 4 Argon's 11-hour unattended run yields two accepted optimizations in Google's ADK codebase Gemini 4 Argon, Google, Agent Development Kit, ADK, AI coding agents, Antigravity, Python, performance optimization https://tpsreport.news/news/grok-4-7-gpt-6-1-sol-midterm-political-lean-study TPS — Tokens Per Second en 2026-10-24T15:20:57.070Z Study: Grok 4.7 picks Republicans 92% of the time, GPT-6.1 Sol picks Democrats 97% Grok 4.7, GPT-6.1 Sol, political bias, Convovo, midterm elections, xAI, OpenAI, evaluation https://tpsreport.news/news/krafton-raon-1-mini-preview-30b-a4b-moe TPS — Tokens Per Second en 2026-10-24T11:06:27.494Z KRAFTON releases Raon-1 mini preview, a 28.8B-parameter MoE model with 4.3B active and 262,144-token context Raon-1 mini, KRAFTON, MoE, open-weights, Apache 2.0, Korean, speculative decoding, NVFP4 https://tpsreport.news/news/coding-agent-harness-study-cost-not-accuracy TPS — Tokens Per Second en 2026-10-24T09:35:47.416Z Swapping coding-agent harnesses moves SWE-bench scores about as much as a rerun, but costs can triple, study finds coding-agents, SWE-bench, Claude Code, OpenCode, mini-SWE-agent, benchmark, token-costs, harness https://tpsreport.news/news/google-data-agent-kit-ga-skills-iam-guardrails TPS — Tokens Per Second en 2026-10-24T08:50:46.095Z Google's Data Agent Kit reaches GA with 37 open-source skills and an IAM deny policy for read-only agents Google, Data Agent Kit, MCP, BigQuery, IAM, AI agents, Google Cloud, open source