6 – 12 July 2026

A blog compared token usage between Claude Code and OpenCode on identical tasks. The trigger was a simple observation: after switching from OpenCode to Claude Code, the usage counter climbed noticeably faster.

To check the hunch, the authors dropped logging between the agent and Anthropic's endpoint and captured every request along with its usage blocks. Their numbers: Claude Code fires off around 33k tokens before it reads the actual prompt, while OpenCode gets by on roughly 7k.

They pin the gap on Claude Code's less efficient caching strategy and the harness's heavier overhead of housekeeping tokens. There's one caveat they add near the end of the post. The Hacker News thread drew 46 points and 16 comments.

Claude Code has a new /checkup command that walks your project and runs 7 maintenance steps. The main one: clearing out unused skills, MCP servers, and plugins to save context.

From there it diffs your local CLAUDE.md against the repo version and strips duplicates, splits a bloated root CLAUDE.md into nested files and skills, and disables slow hooks. The rest is housekeeping: updating Claude Code to the latest version, turning on auto mode by default, and pre-approving read-only commands you keep rejecting.

/checkup asks for confirmation before making any changes. It joins the recently shipped /rewind and /usage.

Anthropic is keeping Fable 5 available on every paid tier and holding Claude Code's weekly limits 50% above normal. Both run through July 19.

You can spend up to half your weekly usage limit on Fable 5. After that, Fable 5 keeps running on usage credits, or you can switch to another model within what's left of your limit.

Anthropic laid out the details in a support article dated July 19.

Vercel added Claude Fable 5 to its v0.app AI generator, alongside Opus 4.8. New accounts receive $5 in monthly credits without a linked card, and referral links grant $5 to both parties.

The referral cap is roughly 40 invites per month, totaling $200 per account. A market has emerged where accounts are registered with temporary emails, verified via phone number, loaded with $200 in credits, and sold for approximately $50.

Users can replicate this process manually. Register on v0.app, verify with a phone number, and use referral links to reach the credit cap. Buying pre-farmed accounts is discouraged, as they rely on the same referral mechanism.

A user reports that a model called Claude Honeycomb appeared in Cursor for a short window before being pulled. He got through just two prompts and posted the results on his profile.

The logs showed a safety fallback to Opus 4.8. His read: fallbacks usually route to a weaker model, so Honeycomb might be stronger than Opus 4.8. Anthropic hasn't confirmed the name, and there's a naming trap here. Per claude.com, Claude already has an official Honeycomb connector for reading and writing observability data, which is a separate thing from any hypothetical model.

Beyond those two Cursor prompts, there's no verifiable data on what Honeycomb actually is.

Anthropic announced J-Space, a feature where Claude performs reasoning steps without displaying them in the final output. The company shared the news via its official account.

No further technical details are available at this time.

New research from Anthropic describes a structural division within Claude that mirrors human cognition. Only a small fraction of the model's internal processes are available for explicit description and reasoning, while the remainder operates outside of conscious access.

Anthropic Academy has pulled together a set of free resources for working with Claude, from API guides to enterprise rollout playbooks. The material is split into three tracks:

• Build with Claude: building apps on the API, with guides and best practices • Claude for work: rolling out Claude across an organization and boosting team productivity • Claude for personal: using Claude on personal projects and everyday tasks

There's also a separate AI Fluency newsletter with research and frameworks, sent once a quarter.

Former Federal Reserve chair Ben Bernanke has been appointed to Anthropic's long-term benefit trust, the body meant to keep the company aligned with its public mission.

Bernanke led the Fed through the 2008 global financial crisis and the recovery that followed, and he's now a senior fellow at the Brookings Institution. Before the Fed, he chaired the economics department at Princeton and later won the Nobel Prize for his research on the Great Depression.

A house at 160 Noe Street in San Francisco is listed at $2,995,000, and the seller is willing to take OpenAI or Anthropic shares as payment, according to ktvu.com and realtor.com.

Per therealdeal.com and abc7news.com, other Bay Area sellers have started asking for pre-IPO stock in AI companies in exchange for multimillion-dollar properties. Demand for the arrangement got wide coverage in June 2026.