https://oh-my-tokens.ai/AI token intelligence for builders, agents, and small teams. Multilingual home pages: /zh/, /pt/, /fr/, /ja/, /ko/.https://simonwillison.net/2026/Jul/3/judgement/https://simonwillison.net/2026/Jul/3/judgement/Fri, 03 Jul 2026 00:00:00 GMTAgent savings can come from fewer bad loops, not only cheaper tokens. Agent 省钱不只是少 token，而是少干预、少返工、少错误 loop。https://simonwillison.net/2026/Jul/2/llm-coding-agent/https://simonwillison.net/2026/Jul/2/llm-coding-agent/Thu, 02 Jul 2026 00:00:00 GMTA minimal coding agent maps where token spend happens. 一个 coding agent 到底由哪些 token-consuming steps 组成？https://simonwillison.net/2026/Jul/2/dspy-datasette-agent-prompts/https://simonwillison.net/2026/Jul/2/dspy-datasette-agent-prompts/Thu, 02 Jul 2026 00:00:00 GMTPrompt optimization can be evaluated with harnesses instead of vibes. 少花 token 的第一步：知道你的 system prompt 是否真的有效。https://www.latent.space/p/vercel-agents-new-softwarehttps://www.latent.space/p/vercel-agents-new-softwareFri, 03 Jul 2026 00:00:00 GMTAgent-readable websites are becoming part of the product surface. 未来网站不只是给人看，也要给 agent 读。https://www.latent.space/p/cursor-forward-deployed-engineershttps://www.latent.space/p/cursor-forward-deployed-engineersWed, 01 Jul 2026 00:00:00 GMTVibe coding becomes a team budget problem when workflows scale. Cursor 的真正成本不是订阅费，而是整个软件工厂的 token burn。https://simonwillison.net/2026/Jun/30/claude-sonnet-5/https://simonwillison.net/2026/Jun/30/claude-sonnet-5/Tue, 30 Jun 2026 00:00:00 GMTNew model releases affect defaults, agent costs, and failure rates. 模型升级后，是否应该切默认模型？看 docs，不看营销。https://huggingface.co/blog/ibm-research/scarfbenchhttps://huggingface.co/blog/ibm-research/scarfbenchTue, 30 Jun 2026 00:00:00 GMTAgent benchmarks are moving toward real enterprise migration tasks. 真正有用的 agent benchmark 应该衡量任务完成成本。https://simonwillison.net/2026/Jun/29/ornith/https://simonwillison.net/2026/Jun/29/ornith/Mon, 29 Jun 2026 00:00:00 GMTOpen-weight coding models can change the API cost equation. 如果 open models 足够会写代码，个人 agent 成本结构会变。https://www.latent.space/p/ainews-openai-reports-median-internalhttps://www.latent.space/p/ainews-openai-reports-median-internalFri, 26 Jun 2026 00:00:00 GMTOutput token growth is a major hidden cost in agent workflows. Agent 时代，贵的不一定是 input，可能是 output 和 loop。https://www.interconnects.ai/p/glm-52-is-the-step-change-for-openhttps://www.interconnects.ai/p/glm-52-is-the-step-change-for-openMon, 22 Jun 2026 00:00:00 GMTChina/open models are part of global agent cost/performance comparisons. 中国模型不是边缘信息，而是全球 agent cost/performance 版图的一部分。https://huggingface.co/blog/is-it-agentic-enoughhttps://huggingface.co/blog/is-it-agentic-enoughThu, 18 Jun 2026 00:00:00 GMTYour own tooling may matter more than public leaderboard rank. 别问哪个模型最好，问哪个模型在你的 agent stack 上最便宜地成功。https://aider.chat/2025/05/08/qwen3.htmlhttps://aider.chat/2025/05/08/qwen3.htmlThu, 08 May 2025 00:00:00 GMTA durable bridge between China models and coding-agent evaluation. 中文/中国模型在 coding agent 里的位置，应该用实际 coding benchmark 讨论。https://code.claude.com/docs/en/prompt-cachinghttps://code.claude.com/docs/en/prompt-cachingSat, 04 Jul 2026 00:00:00 GMTPrompt caching directly changes speed and token cost. 你以为 Claude Code 慢/贵，其实可能是 cache miss。https://www.ianwootten.co.uk/2026/07/01/ditching-claude-for-opencode-and-openrouter/https://www.ianwootten.co.uk/2026/07/01/ditching-claude-for-opencode-and-openrouter/Wed, 01 Jul 2026 00:00:00 GMTA real switching case from default tools to open router/model workflows. 什么时候值得离开默认工具，改用开放 router 和开源模型？https://contextify.sh/https://contextify.sh/Sat, 04 Jul 2026 00:00:00 GMTAgent history and reusable context can reduce repeated token spend. 如果每次 agent 都忘记上下文，你就在重复烧 token。