Freshness Surface

Claude Builds Freshness Log

This page is designed to be updated weekly. It records what changed in the gallery, which sources were checked, and which leads need verification before inclusion.

Updated 2026-09-197 minClaude Code changelogfreshnessbuild log

Key Facts

Page summary for citation

  • Freshness Surface: This page is designed to be updated weekly. It records what changed in the gallery, which sources were checked, and which leads need verification before inclusion.
  • Evidence trail: 10 sources, including Anthropic Docs, Anthropic, Anthropic.
  • Internal context: related pages include Gallery, Venus Map, GPU Kernels.

September 19, 2026 build notes

Two build notes were added from the Claude Fable 5.1 and Claude Mythos 5.1 announcement published on 2026-09-01. Both are vendor-reported research artifacts rather than customer case studies, and both pages label that limit.

The first is a high-resolution elevation map covering a third of Venus, which Anthropic reports Claude Fable 5.1 produced from NASA Magellan radar data and released under a Creative Commons license. The second is a set of custom GPU kernels that Anthropic reports Claude Mythos 5.1 wrote for seven open-source protein and genomics models, with identical outputs and speedups of up to 2.5 times.

Nothing else was added. No new customer story or public repository cleared the verification rubric during this pass, so the rest of the work was editorial: model and surface attribution on every build card, a new browser and desktop surface pattern, and a rule for naming models without overclaiming.

  • Added: /builds/venus-elevation-map, dated to the 2026-09-01 announcement.
  • Added: /builds/gpu-kernel-speedups, dated to the same announcement.
  • Changed: every build card now states its Claude surface, including when the source names no model.
  • Changed: verification rubric grew a model-generation question and a naming rule.

Current edition

The gallery still leads with the builds that have the strongest public source material: Zettair, Spotify, Ramp, Rakuten, official Claude Code build surfaces, and public GitHub templates. The two September research artifacts sit below them because a vendor announcement is weaker evidence than a customer case study or an inspectable repository.

Future updates should add new builds only when they pass the verification rubric. This page keeps the update surface visible for search engines and readers without turning the site into a general news digest.

Watchlist lanes

Good future additions will likely come from official customer stories, public repositories with explicit Claude Code attribution, technical blog posts by builders, and projects surfaced by the broader Claude Code community.

The watchlist should avoid cannibalizing claudeweekly.com. News can be used as a lead, but this site should only publish once a build has enough durable evidence to be useful after the news cycle ends.

  • Publication of the GPU kernel optimizations, which would move that note from reported to inspectable.
  • Use of the Venus elevation map by the NASA VERITAS or ESA EnVision teams.
  • The first customer story or public repository that names a model generation, which would let a build card cite one.
  • Public artifacts built in Claude Cowork or Claude in Chrome with an inspectable output rather than a screenshot.

How submissions become updates

The submission formatter turns a candidate build into the same shape the editor needs: artifact URL, evidence URL, Claude surface, artifact type, current status, evidence summary, and limits. That keeps future additions comparable to the first-edition build cards.

A future update should re-check the public source, decide whether the candidate belongs in the gallery or only the watchlist, add the source to the ledger, and link the new page from at least three related pages. That sequence keeps freshness useful without lowering the verification bar.

Update Log

Source passes

September 19, 2026 magic pass

Added The Lightbox, a signature plate gallery with evidence trails for sourced artifacts, including high-level Fable 5.1 / Mythos 5.1 science examples.

  • Added /lightbox/ as the characteristic experience: click a plate for artifact, claim, license, and independent-reproduction criteria.
  • Spotlit the Venus DEM, GPU kernels, and Mythos protein-binder experimental validation at announcement altitude; no dual-use methods.
  • Kept Zettair, Spotify, Ramp, Rakuten, GitHub Actions, and Anthropic internal plates without restamping unnamed model generations.

September 19, 2026 source pass

Model and surface attribution added across the gallery, plus two vendor-reported research-artifact build notes from the Claude Fable 5.1 announcement.

  • Added the Venus elevation map build note, sourced to the 2026-09-01 Fable 5.1 announcement.
  • Added the GPU kernel speedup build note from the same announcement.
  • Every build card now names its Claude surface, including when the source names no model generation.
  • Added a browser and desktop surface pattern covering Claude in Chrome and Claude Cowork.
  • Verification rubric grew a model-generation question and a rule against relabeling older metrics.

July 6, 2026 source pass

First edition: the gallery, the verification rubric, and the source ledger were published.

  • Initial gallery with Zettair, Spotify, Ramp, Rakuten, official build surfaces, and public repositories.
  • Established the evidence tiers that later passes have to keep using.

Primary Sources

Source trail

Links are cited for factual claims on this page. Access dates are kept so future updates can re-check drift.

Claude Code overview

Anthropic Docs - official docs - accessed 2026-07-06

Open citation

Current capability overview: local, web, IDE, git, MCP, hooks, skills, and agents.

Claude Opus 5

Anthropic - official blog - accessed 2026-09-19

Open citation

Opus 5 launch on 2026-07-24, the default on Claude Max and the strongest model on Claude Pro.

Choosing a Claude model

Anthropic Docs - official docs - accessed 2026-09-19

Open citation

Current generally available lineup and official guidance on picking between Fable 5.1, Opus 5, Sonnet 5, and Haiku 4.5.

Ramp Claude Code case study

Anthropic Customer Stories - official case study - accessed 2026-07-06

Open citation

Ramp engineering workflows for test automation, docs, parallel development, incidents, and ticket-to-code.