Sunday, July 26, 2026

Several pricing adjustments were observed, with Claude Code's typical monthly cost changing from $17 to $20/mo (https://claude.com/pricing). Cursor's floor price shifted from $20 to $0/mo (https://cursor.com/pricing), while Kiro's floor price moved from $0 to $20/mo (https://kiro.dev/pricing). Windsurf's floor price also changed from $20 to $0/mo, and its model shifted from hybrid to flat (https://devin.ai/pricing).

Discussions among the community included agentic test processes, LLM benchmarks, and other notes on agentic coding (https://news.ycombinator.com/item?id=49054268). There was also conversation about the new rules of context engineering for Claude 5 generation models (https://news.ycombinator.com/item?id=49051361, https://news.ycombinator.com/item?id=49046425), and how to compress context in Claude Code without losing details (https://news.ycombinator.com/item?id=49048571).

New tools were presented, such as Awsmux, a multi-account AWS CLI that claims to be up to 5.4x faster and use up to 7.4x fewer tokens for agents (https://news.ycombinator.com/item?id=49051043). HotPin was also introduced, enabling lossless 120B MoE inference on 24GB RAM via CPU, tested with qwen, glm, and llama models (https://news.ycombinator.com/item?id=49050356). A question was raised regarding what individuals do when their AI agents are working (https://news.ycombinator.com/item?id=49049986). Separately, OpenAI Codex experienced downtime, which is currently under investigation (https://news.ycombinator.com/item?id=49046018).