Welcome back to Agentic Coding Weekly. Here are the updates on agentic coding tools, models, and workflows for the week of Jun 28 - July 4, 2026.
Executive Summary:
Anthropic released Claude Sonnet 5. Scores 63.2% on SWE-bench Pro and 80.4% on Terminal Bench 2.1, similar to GLM 5.2 but much more expensive.
Fable 5 is back. Available as part of Claude subscriptions until July 7th.
Zai launched, Zcode, a new Agentic Development Environment (ADE), similar to Codex desktop, for GLM 5.2.
GPT 5.5 bug causes it to produce exactly 516 or 1034 or 1552 reasoning tokens and a degraded output.
In a new Claude Code feature, model asks questions using the question tool and if we don’t reply within 60 seconds, it continues without an answer.
Cursor launched their ios app.
Worth reading: local development with Qwen 3.6 27B, Claude Code marking requests by sneakily changing date string in system prompt, and a concrete example of working with AI.
1. Tooling and Model Updates
Claude Sonnet 5
Sonnet 5 is upgraded version of Sonnet 4.6. The model has a new tokenizer, where the same input text produces approximately 30% more tokens than on Claude Sonnet 4.6 which means effectively a 30% price increase.
Scores 63.2% on SWE-bench Pro and 80.4% on Terminal Bench 2.1, similar to GLM 5.2 but much more expensive. Compared with Opus, it's 1.1% behind Opus 4.7 on SWE-bench Pro and 14.3% ahead of Opus 4.7 on Terminal Bench 2.1.