Builder source library

Run ready projects. Inspect every classified output.

One Builder membership unlocks the verified Launch 002 Project Pack, its immutable Research Pack, and Tech Review 001 Research Pack. Future published Builder project packs are included while membership is active; no fixed cadence is promised.

Versioned v1 release33 classified projectsMissing evidence stays labeled
Project outputs
33
Blender scenes
14
Browser + shaders
11 + 8
Full archive
481.0 MB
Currently published · one entitlement

Choose what you want to run or inspect.

Launch 002 has separate Project Pack v1 and Research Pack v1 downloads. Tech Review 001 remains a Research Pack. Every published release is immutable and independently checksum-verified.

Tech Review 001 · v1

  • 7 exact promptsGPT-5.6 vs Claude Fable 5
  • 28 run ledgers4 disclosed configurations
  • 44 verified filesExports, matrix, manifests, checksums, README, and VERSION
Open collection

Kimi K3 Launch 002 · v1

  • 33 reviewed outputs14 Blender · 11 browser · 8 shader
  • Readiness is explicit25 ready · 5 reference-only · 3 partial · 0 excluded
  • Full pack + 9 task packs481.0 MB full archive · runtime notes and licenses included
  • 9 exact promptsKimi K3 vs GPT-5.6 vs Claude Fable 5
  • 33 run ledgers6 disclosed configurations
  • 51 verified filesExports, matrix, manifests, checksums, README, and VERSION
Open source collection
Free preview · 1 of 9

Inspect an exact prompt and its matching ledger.

This reviewed sample is copied from the pinned Kimi K3 Launch 002 · v1package. The raw ledger remains protected; only these sanitized fields are shown here.

Exact frozen prompt

Campfire Under a Starry Night

v1

Make a camp fire under a starry night in threeJS

This frozen prompt does not require a task-specific input asset.

Sanitized ledger sample

Same Campfire Under a Starry Night target

Public receipt
Model / configuration
Kimi K3 · MaxKimi Code CLI
Reported cost
$0.45API-equivalent usage accounting, not an itemized subscription cash charge
Processed tokens
191,230
Workflow time
14:19.3End-to-end
Result status
Partial token, timing and source ledger
Open the public JSON receipt
Available now · Kimi K3 vs GPT-5.6 vs Claude Fable 5

Inside Kimi K3 Launch 002 v1.

Two separate immutable deliverables: Project Pack v1 for source artifacts and Research Pack v1 for prompts, ledgers, exports, and evidence controls. Both are entitlement-checked and checksum-verified.

Project Pack v1

  • 33 classified project outputs25 ready · 5 reference-only · 3 partial · 0 excluded
  • 14 editable sanitized Blender scenesDerived customer copies with disclosed runtime notes
  • 11 browser projectsCustomer-safe source trees, lockfiles, permitted assets, and RUN notes
  • 8 shadersShader source and packaged runner material
  • One full archive + 9 task archives481.0 MB for the complete download
  • Licenses, notices, checksums, and manifestsUse only where the included terms grant it

Research Pack v1

  • 9 exact promptsFrozen benchmark instructions in Markdown
  • 33 available raw ledgersCost, token, timing, status, caveats, and evidence gaps
  • 6 configuration recordsModel, reasoning level, and provider agent
  • Aggregate and per-run exportsCost, processed-token, and workflow-time tables
  • Missing-evidence matrixNo blank field is silently promoted to proof
  • Version manifest, checksums, and READMEVerify the 51-file release
Archive structure

remakebench-kimi-k3-launch-002-v1.zip

51 files
  • prompts/9 × .md
  • results/33 × .json
  • configurations/configurations.json
  • exports/2 × .csv · aggregate + per-run
  • manifests/2 × .json · public + private
  • missing-evidence-matrix.csv1 file
  • checksums.sha2561 verification file
  • README.md + VERSION1 README + release version

This archive is the separate immutable Research Pack v1. Download Project Pack v1 for the approved source projects and editable output artifacts.

Use the records

What you can do with the package.

  1. 01

    Run ready projects; inspect every output

    Launch the 25 ready projects with their disclosed runtimes. Inspect the 5 reference-only and 3 partial outputs with their per-output status reasons and limitations; no runnable claim is made for them.

  2. 02

    Audit workflow economics

    Break down reported cost, processed tokens, and end-to-end time by run instead of relying on a headline total.

  3. 03

    Compare 6 configurations

    Inspect how model, reasoning level, provider agent, caching, and workflow shape differ across the 33 records.

  4. 04

    Trace the evidence boundary

    Use the missing-evidence matrix and checksums to separate what was recorded from what the review does not establish, then bring the receipt to the private #builder-lab reproduction forum.

Free + Builder

See the result free. Run ready projects with Builder.

Public proof

Free

$0

See the result

  • Public results and standardized output media
  • Anonymous blind battles
  • Headline methodology, cost, and workflow time
  • Visible evidence gaps and record status
  • Verified community Discord after account creation
Join the verified member Discord
Full research stack

Builder

$29.99 / month

Audit the workflow behind the result

  • Customer-safe project materials across 33 reviewed Launch 002 outputs
  • 14 editable sanitized Blender scenes, 11 browser projects, and 8 shaders
  • Project status: 25 ready · 5 reference-only · 3 partial · 0 excluded
  • Full Project Pack plus nine task-sized downloads
  • Every currently published exact-prompt research package
  • Complete available run ledgers and disclosed configurations
  • Cost, token, and workflow-time exports
  • Missing-evidence matrices, manifests, and checksums
  • One entitlement across Tech Review 001 and Launch 002
  • Private Builder Lab: #builder-releases and #builder-lab
Before checkout

Exactly what you are buying.

What exactly downloads?

Launch 002 has a 481.0 MB full Project Pack, nine smaller task packs, and a separate 51-file Research Pack. The project downloads contain customer-safe source artifacts, explicit ready/reference-only/partial/excluded classifications, run notes, receipts, manifests, checksums, licenses, and notices; the research download contains the nine prompts, 33 ledgers, configurations, and exports.

Is this every file used to make the video?

No. Project Pack v1 contains only the reviewed output artifacts and inputs whose provenance, redistribution, dependency, security, and runtime gates passed. Private sessions, credentials, local paths, protected evidence, and unresolved third-party bytes remain excluded; each task documents its limitations.

What stays available without Builder?

The comparison, blind play, headline methodology, public cost and workflow-time evidence, and visible evidence gaps remain public. A free account can also join the Member Discord.

How do billing and delivery work?

Builder is billed monthly at $29.99 / month. After Stripe confirms the subscription, Builder access appears in your account and the protected collection download unlocks. Billing is managed through the secure Stripe portal.

Does the subscription promise future review packages?

Builder includes future project or research packs that RemakeBench publishes while your membership is active, but no fixed cadence, volume, or unpublished scope is promised. This page names only releases that have passed their publication gates.

Available now · Kimi K3 Launch 002 · v1

Take the projects and receipts with you.

33 projects. 9 task packs. 9 exact prompts. 33 ledgers. Separate immutable project and research releases.

$29.99 / month