Welcome back to Agentic Coding Weekly. Here are the updates on agentic coding tools, models, and workflows for the week of Jul 5 - 11, 2026.
Executive Summary:
GPT 5.6 family of models, Sol, Terra, and Luna, are now publicly available. Sol is priced similarly to GPT 5.5, $5 / $30 per million input / output tokens.
Grok 4.5 is the new model from xAI, jointly developed with Cursor. 500k context window, similar performance to GLM 5.2, and $2 / $6 per million input / output tokens pricing.
Muse Spark 1.1 is the latest model from Meta. Priced cheaper than GLM 5.2 and Grok 4.5.
Hy3 is a new open-weights model from Tencent. 295B total parameters, 21B active. 256k context window. Supports only text inputs.
xAI's Grok Build CLI sends the whole repository back home independent of what the agent reads. Beware!
Fable 5 will be available on all subscription plan until July 19th.
Claude Code 50% higher weekly limits promotion ending on July 19th.
Worth reading: Why SWE-bench Pro is no longer recommended, how was Bun rewrite in Rust done, and burnout from reading LLM generated texts.
Meanwhile, find out how many words out of 18 you can unscramble, with 30 seconds per word limit.
Non-AI disclaimer: I curate and write every issue, including this one, myself. See my reasoning here.
1. Tooling and Model Updates
GPT 5.6
GPT 5.6 family of models, Sol, Terra, and Luna, are now publicly available. On coding benchmarks, Sol max seems to match or beat Fable 5 max performance. On xhigh reasoning, Sol seems to match/beat GPT 5.5 xhigh and beat Opus 4.8 xhigh.
On Terminal Bench 2.1, Sol max scores 88.8% compared to 88.0% of Mythos 5/Fable 5.
On xhigh reasoning though, which is what I use for coding tasks, GPT 5.6 Sol scores 84.9% on Terminal Bench 2.1 compared to 85.6% scored by GPT 5.5 at the same xhigh reasoning.