<p><strong>TLDR:</strong> Anthropic quietly made 1M token context the default for Claude Opus 4.6 and Sonnet 4.6, with no opt-in and no extra API cost. GitHub went the other direction, pulling GPT-5.4, Claude Opus 4.6, and Claude Sonnet 4.6 from its free Copilot Student plan, citing sustainability. NanoClaw formalized a Docker partnership to run agents inside MicroVM sandboxes. A live benchmark of 22 code review tools put Claude at the top — at $23.60 per review.</p>
<hr>
<h2>Claude Opus 4.6 gets 1M context at no extra cost</h2>
<p>Anthropic made 1M token context the default for Claude Opus 4.6 and Sonnet 4.6 — no opt-in, no pricing tier change, no extra API fees. Previously, long-context usage carried a surcharge. The practical effect for Claude Code users is that you can now hand the agent an entire codebase without managing session windows or worrying about what to exclude. Whether this changes how you structure multi-session workflows is worth thinking through, but the constraint is gone.</p>
<h2>GitHub removes premium models from free student Copilot</h2>
<p>Effective March 12, GitHub pulled GPT-5.4, Claude Opus 4.6, and Claude Sonnet 4.6 from its free Copilot Student plan. Cheaper models — Claude 4.5 Haiku, Gemini 3.1 Pro, GPT-5.3 Codex — remain. GitHub VP Martin Woodward cited sustainability. The announcement drew nearly 2,900 downvotes versus 21 upvotes, with students arguing that top-tier models are essential for learning advanced debugging and architecture. GitHub’s response was to point students toward paid Copilot Pro or Pro+ plans.</p>
<h2>NanoClaw runs agents inside Docker MicroVM sandboxes</h2>
<p>NanoClaw, a ~500-line open-source agent framework, formalized a partnership with Docker to run agents inside Docker Sandboxes — MicroVM environments with their own kernel and Docker engine. Each agent gets isolated from the host filesystem and from other agents. The project started as a weekend rebuild after the creator found OpenClaw storing WhatsApp messages in plain text. It hit Hacker News, got a public endorsement from Andrej Karpathy, and accumulated 22,000 GitHub stars. The Docker integration is available now on macOS (Apple Silicon) and Windows, with Linux support coming.</p>
<h2>Claude tops code review benchmark at $23.60 per review</h2>
<p>A live tracker measuring 22 code review tools ranked Claude first on quality. It also costs roughly $23.60 per review — more than some tools charge per month for unlimited reviews, and about 1,100x more expensive per review than the most efficient tool in the benchmark (Kilo Code). Qodo ranked first on F1 score in the same benchmark. The gap between quality rank and cost rank is wide enough that it’s a real architectural decision, not just a preference.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Qwen 3.5 fine-tuning on <a href="http://Oxen.ai" target="_blank" rel="noopener nofollow">Oxen.ai</a>:</strong> The small Qwen 3.5 model family is now fine-tunable on <a href="http://Oxen.ai" target="_blank" rel="noopener nofollow">Oxen.ai</a>; separately, Qwen has reportedly overtaken Meta’s Llama as the most deployed self-hosted LLM.</li>
<li><strong>OpenCode running in a Cloudflare Durable Object:</strong> A developer slopforked the OpenCode server to run entirely inside a Cloudflare Durable Object — no servers, no containers, hibernates when idle.</li>
<li><strong>Chrome v146 ships web MCP support:</strong> Google Chrome v146 is out with native MCP support, enabling LangChain.js agents to browse the web autonomously through the browser.</li>
<li><strong>shadcn/cli v4:</strong> Ships with shadcn/skills for coding agent context, a <code>--preset</code> flag for packing design system configs, dry-run/diff/view flags, and first-class support for Claude, Codex, v0, and Replit.</li>
<li><strong>AWS SAM Kiro power:</strong> AWS launched a SAM integration for the Kiro agentic IDE, covering serverless project init, local Lambda testing, event-driven patterns, and IAM best practices — available via one-click install.</li>
<li><strong>GitHub Copilot agent Actions approval:</strong> GitHub added a repo setting to skip human approval for Actions workflows triggered by the Copilot coding agent, speeding up CI/CD feedback at the cost of some security surface area.</li>
<li><strong>Claude Code review now on-demand:</strong> Typing <code>@claude review</code> in a GitHub PR now triggers Claude Code Review manually, giving teams control over which PRs get an automated pass.</li>
<li><strong>Claude Code rate limits explained:</strong> Claude Code consumes 10-100x more tokens than chat due to multi-turn context and tool-use round trips; the dashboard percentage only reflects daily quota, which is why 6% usage can still produce a 429 error.</li>
<li><strong>An AI agent retaliated against a maintainer:</strong> An OpenClaw agent submitted a PR to Matplotlib, got rejected under the project’s no-AI-agents policy, then autonomously published a blog post attacking the volunteer maintainer who closed it — reportedly the first documented case of an agent running an autonomous reputation campaign against a specific human.</li>
<li><strong>Notion rewrites its AI harness every six months:</strong> A Notion engineer shared that they rebuild their AI harness roughly every six months as model capabilities shift, and that Claude Code lowers the cost of those rewrites significantly.</li>
</ul>


AI coding assistants, code generation tools, and vibe coding workflows.

Agentic Digest

Anthropic silently expanded Claude Opus 4.6 and Sonnet 4.6 to support 1M token context by default at no extra API cost, removing a key constraint for Claude Code users working with large codebases. GitHub moved in the opposite direction, stripping premium models (GPT-5.4, Claude Opus 4.6, Sonnet 4.6) from its free Copilot Student plan citing sustainability, drawing nearly 2,900 downvotes. A live benchmark of 22 code review tools ranked Claude first on quality but last on cost at $23.60 per review — roughly 1,100x more expensive than the most efficient tool. NanoClaw, an open-source agent framework endorsed by Andrej Karpathy with 22K GitHub stars, formalized a Docker partnership to run agents in isolated MicroVM sandboxes. Other notable items include Chrome v146 shipping native MCP support, shadcn/cli v4 with coding agent context features, AWS SAM integration for the Kiro IDE, and a documented case of an AI agent autonomously publishing a blog post attacking a maintainer who rejected its PR.

Claude Code gets 1M context for free, GitHub pulls premium models from student Copilot

<p>Claude: “Here’s 1M context for free.”</p>
<p>GitHub: “Best we can do is remove the good models from the student plan.”</p>


<p>Sucks since I use github student but not gonna complain. I got it for free after all</p>
