林悦己AI × CONTENT × GROWTH

RECENTLY UPDATED / (SHANGHAI TIME)

LIN YUEJI / ARCHIVE

GPT-6 Sol & Luna Use Cases

Explore public GPT-6 Sol and Luna cases across web builds, games, agents, video workflows, and evaluations, with model roles and reported limits made clear.

58 REAL CASES6 CATEGORIES3 LANGUAGES
Browse model catalog ↗58 GPT-6 Sol & Luna use cases ↓BACK TO RESOURCES

USE CASE MAP / 01

Real ways to use GPT-6 Sol & Luna, beyond a capability sheet

Only public tasks, evaluations, and failures clearly attributed to GPT-6 Sol or Luna are included. Each case links to its creator and source; most results are author or platform reports, not independent reproductions.

58curated real-world cases
6task categories
3English, Chinese, and Japanese

HOW TO USE / 02

From case discovery to workflow validation

Check the model variant and task on each card, then open the source for implementation details, tool dependencies, and evidence limits.

  1. 01

    Browse by task

    Start with coding, agents, creative work, evaluation, or another task family.

  2. 02

    Search the outcome

    Use titles, methods, or creators to find the closest case to your current goal.

  3. 03

    Verify the source

    Select the creator to review the full demo, limits, context, and original explanation.

  4. 04

    Reproduce on EvoLink

    Choose a method worth testing, connect to the model, and turn it into your workflow.

REAL CREATOR CASES / 03

58 GPT-6 Sol & Luna use cases

Up to four cases per row. Select a creator to open the original case; images and videos load only when needed.

FILTER BY CATEGORY

58 cases

CASE 01Evaluation
Web, 3D & SVG

Sol and Luna in a landing-page prompt comparison

A creator added both GPT-6 models to a same-prompt landing-page gallery; the published results are a comparison, not an independently reproduced build.

@nemumusitocha
CASE 02Evaluation
Web, 3D & SVG

Sol: Build a mechanical hummingbird in Three.js

The creator compared Sol with the earlier Opus 5 using the same prompt and max effort; this is not an Opus 5.5 comparison.

@TimJayas
CASE 03Evaluation
Web, 3D & SVG

Compare Sol and Opus on a 3D creation task

An API platform showed a 3D quality and cost comparison; the result is platform-reported and was not reproduced.

@aimlapi
CASE 04Evaluation
Web, 3D & SVG

Sol: Animate a cycling pelican in Blender

An API platform gave Sol and Opus 5.5 the same brief for a looping 3D pelican animation; the aesthetic verdict is the author’s.

@atomic_chat_hq
CASE 05Evaluation
Web, 3D & SVG

Generate a voxel pagoda with Luna Max

A same-task visual comparison pits GPT-6 Luna Max against GPT-5.6 Luna Max; source media shows the output, without code verification.

@AiBattle_
CASE 06Evaluation
Web, 3D & SVG

Draw a fidget-spinner SVG with Sol xHigh

The post compares a specific SVG task under Sol xHigh and Astra High; visual output is shown, while source code remains unverified.

@SadAlbert10
CASE 07Limit
Web, 3D & SVG

Sol: New York Three.js scene stops at a static frame

The creator reports that Sol produced a static frame without the requested render loop or camera motion; this is a reported incomplete result.

@aipulseda1ly
CASE 08Evaluation
Games & interactive prototypes

Sol: Build a 3D game in Unreal Engine

Higgsfield compared Sol and Opus 5.5 in an Unreal Engine game build; the project and playability were not independently tested.

@higgsfield_ai
CASE 09Demo
Games & interactive prototypes

Make a Mario-style game with Luna Max

The creator shows a short Luna Max game demonstration; the game was not independently run.

@benchmark_lb900
CASE 10Demo
Games & interactive prototypes

Make a Minecraft-style game with Sol

The creator demonstrates a Minecraft-style build attributed to Sol; source code and playability were not verified.

@karankendre
CASE 11Demo
Games & interactive prototypes

Sol: Create the Neon Exit escape game

JetBrains showcased a Sol-assisted Neon Exit game made by an employee; this is a platform-published demonstration.

@jetbrains
CASE 12Demo
Video tool workflows

Produce a short Luna introduction video

A creator says Luna made a model introduction video; the production chain is undisclosed and does not establish native video generation.

@nemumusitocha
CASE 13Evaluation
Video tool workflows

Compare Tesseract video workflows with Sol and Astra

Mirage showed Tesseract-assisted results for Sol and Astra; the rendering role belongs to the video tool, and the comparison was not reproduced.

@trymirage
CASE 14Demo
Video tool workflows

Sol: Create a video through HyperFrames

A creator demonstrated a few-prompt workflow combining Sol with HyperFrames; the post does not show native Sol video generation.

@Miguel07Code
CASE 15Demo
Video tool workflows

Sol: Rebrand and resize a video with Tesseract

Tesseract’s co-founder shows Sol-assisted brand and aspect-ratio edits; the result is a product demonstration, not independent validation.

@dwightchurchill
CASE 16Benchmark
Agents, coding & browser tasks

Compare Sol and Luna on long browser tasks

Browser Use published a benchmark of long, difficult browser-agent tasks; its cost and performance claims are platform-reported.

@gregpr07
CASE 17Integration
Agents, coding & browser tasks

Move scheduled tasks from Astra to Sol

The author reports switching existing scheduled tasks to Sol; replacing another API with Luna remained a future plan.

@gengdaJ
CASE 18Evaluation
Agents, coding & browser tasks

Sol + Luna: Test tool use with ts-bench

A researcher reports improved Sol tool use and weaker Luna performance relative to the prior generation; the evaluation was not reproduced.

@gosrum
CASE 19Benchmark
Agents, coding & browser tasks

Sol: Run Next.js framework evaluation tasks

The Next.js team reports a 97% success rate for Sol on its framework eval; this does not measure all software development tasks.

@nextjs
CASE 20Evaluation
Agents, coding & browser tasks

Luna: Run a full agent end-to-end test suite

A developer reports both suites passing with Luna at $0.03 versus $0.15 for the prior model; these are author-recorded costs.

@o_kwasniewski
CASE 21Demo
Agents, coding & browser tasks

Sol: Draw a portrait in browser Paint

A creator instructed Sol to open Paint in the browser and draw a portrait; the post shows the reported result.

@The_Alex
CASE 22Evaluation
Research & writing

Sol: Write and blind-review YouTube scripts

A publisher asked models for 50 scripts across ten genres and used three AI judges; Sol’s reported result reflects this test, not audience response.

@Whats_AI
CASE 23Evaluation
Research & writing

Evaluate Sol for wide-and-deep research

Perplexity reports a WANDR research evaluation and plans Sol for its Light orchestrator; planned routing is separate from measured results.

@AravSrinivas
CASE 24Limit
Robotics & limitations

Luna fails a LIBERO robot manipulation task

A researcher reports Luna failing to open a stove and place a moka pot; the task name comes from a quoted Astra experiment.

@YuXiang_IRVL
CASE 25Evaluation
Web, 3D & SVG

Sol + Luna: build a lava lamp

BridgeBench compared four models on one lava-lamp prompt; the posted Sol/Luna times and costs are test-specific.

@bridgebench
CASE 26Benchmark
Agents, coding & browser tasks

Sol: find hidden bugs in two repositories

The publisher reports Sol Max finding 29.3 of 105 seeded bugs at a lower estimated cost than several peers; quality fell in this test.

@PawelHuryn
CASE 27Limit
Agents, coding & browser tasks

Luna: repo review takes the wrong tool path

The author says Luna ignored a GitHub integration and opened a browser instead during a repository review, across four harnesses.

@Da7_Tech
CASE 28Demo
Games & interactive prototypes

Sol: prototype a dogsled racing game

Higgsfield showed a Sol versus Opus 5.5 dogsled game comparison; the platform supplied the creation environment.

@higgsfield_ai
CASE 29Evaluation
Web, 3D & SVG

Sol: build a sunset-ocean scene

BridgeBench reports Sol completing the shared scene prompt in 44 seconds at $0.07; this is one visual comparison.

@bridgebench
CASE 30Demo
Games & interactive prototypes

Sol: prototype a 3D strategy game

Higgsfield compared Sol and Opus 5.5 on a battlefield-scale game prototype; the result is a platform demo.

@higgsfield_ai
CASE 31Demo
Games & interactive prototypes

Sol: build a multiplayer forest board game

A creator reports using Sol with Combos CLI for a 2–4 player board game; the external tool supplied assets, networking, and publishing.

@cnyzgkc
CASE 32Benchmark
Research & writing

Luna: test visual extraction and counting

Roboflow reports lower extraction, counting, and reasoning scores than GPT-5.6 Luna, while detection improves slightly.

@skalskip92
CASE 33Limit
Web, 3D & SVG

Luna: peacock animation uses a visual shortcut

In a peacock animation test, the creator says Luna inserted an image and spinning wheel circles instead of constructing the requested animation.

@SahilPanhotra
CASE 34Evaluation
Games & interactive prototypes

Sol + Luna: build a responsive cursor pet

The prompt asked for a monster that follows the cursor and reacts to feeding; the publisher compared both GPT-6 variants and reported Luna finishing first.

@noclipepe
CASE 35Evaluation
Games & interactive prototypes

Sol + Luna: build a neon pinball game

A same-prompt test compared playable pinball builds; the publisher reports Luna at 6:22 and Sol at 17:26.

@noclipepe
CASE 36Demo
Video tool workflows

Sol: code a looping water-cycle animation

Higgsfield reports a single HTML project with browser-rendered frames and timeline controls; Sol’s role was code generation.

@higgsfield_ai
CASE 37Evaluation
Web, 3D & SVG

Sol: build an interactive planets website

The creator reports a three-planet site from Sol in ten minutes; the comparison used a different subscription plan for Opus.

@Kappaemme1926
CASE 38Evaluation
Web, 3D & SVG

Sol: design a kanban web app

A creator compared four models on a Hermes Agent kanban app prompt and ranked Sol last for frontend design.

@viticci
CASE 39Evaluation
Web, 3D & SVG

Sol: render an ancient Chinese city in 3D

OpenDesign compared Sol and Opus 5.5 on the same city prompt and reported distinct visual styles.

@OpenDesignHQ
CASE 40Benchmark
Agents, coding & browser tasks

Sol + Luna: rediscover known vulnerabilities

A 32-CVE benchmark reports 68.8% recall for Sol and 53.1% for Luna; both trail their GPT-5.6 predecessors on recall.

@pilvar222
CASE 41Limit
Web, 3D & SVG

Sol: Blender scene shows rigging flaws

A same-prompt Blender test reports faster, cheaper Sol output with visible geometry and rigging problems.

@alwayspriyesh
CASE 42Benchmark
Agents, coding & browser tasks

Sol + Luna: test 100 coding environments

GertLabs reports Sol near Opus 5.5 within its margin of error and Luna slightly ahead of 5.6 Luna in its custom coding evaluation.

@leo_linsky
CASE 43Demo
Games & interactive prototypes

Sol: build a multiplayer 3D arena game

The creator used Astra for planning and Sol for coding, with Combos CLI and other models for 3D assets, sound, multiplayer, and publishing.

@Saccc_c
CASE 44Demo
Agents, coding & browser tasks

Sol: discuss a YouTube video while it plays

A creator built a co-watch site using ChatGPT Voice and WebMCP; Sol comments on a synced video during playback.

@miu21590
CASE 45Evaluation
Research & writing

Sol: draft a shared-runners product post

The same prompt asked four models to write a product update with a UI screenshot; the author estimated Sol’s cost at $3.10 and preferred another result.

@thorstenball
CASE 46Benchmark
Research & writing

Sol: attempt a 3D spatial maze benchmark

MazeBench reports Sol scoring 1% on its difficult spatial tasks; this is a narrow benchmark result.

@patience_cave
CASE 47Benchmark
Web, 3D & SVG

Sol: rank in Code Arena WebDev

Arena reports Sol Max at fourth place in its WebDev ranking; the score reflects that evaluation’s prompts and voters.

@arena
CASE 48Benchmark
Agents, coding & browser tasks

Sol + Luna: run Rails agent tasks

A Rails agent evaluation reports Luna Max completing 18% of tasks for $11 and compares it with Sol in the same chart.

@dhh

Showing 48 / 58

BUILD WITH EVOLINK / 04

Find models for your own validation

Check the current model catalog and integration options on EvoLink, then test a relevant case on your own task.

Browse model catalog ↗