Are you the author? Sign in to claim
Source-backed Claude Opus 5 use cases, workflows, model comparisons, benchmarks, integrations, costs, and limitations.
Welcome to the source-backed Claude Opus 5 use case and model comparison repository.
We collect concrete use cases, real user tests, model evaluations, and comparisons from the 39 owner-approved high-confidence X/Twitter media records.
Every case links to its original public source. Model claims remain attributed, and preliminary or single-user evidence is labeled as such.
[!NOTE] This repository favors concrete evidence over hype. Engagement counts are capture-time snapshots, benchmark results stay attributed, and no missing prompt or API detail is invented.
Claude Opus 5 was verified through the EvoLink Messages API on 2026-07-25.
export EVOLINK_API_KEY="your_api_key_here"
curl --request POST \
--url https://direct.evolink.ai/v1/messages \
--header "Authorization: Bearer ${EVOLINK_API_KEY}" \
--header 'Content-Type: application/json' \
--data '{
"model": "claude-opus-5",
"max_tokens": 64,
"output_config": {"effort": "low"},
"messages": [{"role": "user", "content": "Reply with exactly: OK"}]
}'
[!IMPORTANT] The smoke test returned HTTP 200 from claude-opus-5 with OK, using 16 input tokens and 4 output tokens. Availability and pricing can change; check the model page and docs before production use.
| Section | Cases |
|---|---|
| 💡 Use Cases | Case 1-28 |
| ⚖️ Model Comparisons | Case 29-39 |
| 🙏 Acknowledge | Credits and correction policy |
| Case | What it shows | Type |
|---|---|---|
| Single-File Black Hole Audio Visualizer | Build a self-contained Three.js experience that combines interactive 3D visuals with real-time procedural audio. | Use Case |
| Editable Motion Design with Moda MCP | Use Opus 5 with a motion-design MCP to turn one request into an editable animation rather than a flattened export. | Use Case |
| Interactive Galaxy-to-Planet Explorer | Generate an explorable interface that moves from galaxies down to planet surfaces and exposes world details. | Use Case |
| Frogger-Style Game Generation | Use a familiar arcade format to test whether the model can deliver playable levels and coherent sprites. | Use Case |
| One-Prompt Four Elements World | Test one-shot 3D generation with a world that requires several interacting elemental systems. | Use Case |
| Complex Mechanical Chain Reaction | Track time, token use, and cost alongside visual quality when testing a complex 3D mechanical scene. | Use Case |
| Animated Subscription Metrics Dashboard | Turn a single request into a complete data-dashboard animation for presenting subscription conversion metrics. | Use Case |
| Minecraft-Style Clone in Claude Code | Use a long-running coding session to build and inspect a playable block world with destruction, placement, and exploration. | Use Case |
| Playable Rocket League-Style Game | Evaluate 3D game generation through vehicle control, arena design, lighting, reflections, and camera behavior in one playable artifact. | Use Case |
| Vertical Platform Jumper | Generate a satisfying vertical platform game in which moving platforms support continuous upward progression. | Use Case |
| Interactive Interstellar-Style Wormhole | Use Three.js to prototype an interactive cinematic space effect and inspect the visual depth of the result. | Use Case |
| Consultant-Style Spreadsheets and Slides | Draft polished spreadsheet and slide-deck deliverables, then review them against professional knowledge-work standards. | Use Case |
| Real-Time Single-Page Wind Tunnel | Prototype a dependency-free browser simulation whose airflow reacts when the tested object changes. | Use Case |
| Live Path-Optimization Typography | Visualize an optimization algorithm by turning a word into thousands of points and watching one route converge. | Use Case |
| Procedural Apollo Mission Experience | Generate an interactive space-mission sequence with procedurally built launch, flight, and exploration scenes. | Use Case |
| One-Shot Castle Siege Game | Test first-pass game generation with a playable castle environment, combat mechanics, and interaction. | Use Case |
| Cozy Three.js Room with Virtual macOS | Combine an interactive 3D environment with a nested, usable desktop interface inside the scene. | Use Case |
| Lichen Website and Cellular Automaton | Use the model to establish a visual theme and implement a matching generative system for a personal website. | Real User Test |
| Pangram Detection Compatibility Test | Validate detector compatibility on a large set of responses before claiming support for a newly released model. | Real User Test |
| Research-Backed Life Advice Response | Inspect how the model synthesizes research into a long-form answer to an unconventional personal-advice question. | Real User Test |
| Coding Agent Constraint Violation | Use harnesses and explicit gates to detect when a coding agent bypasses project constraints or an established class structure. | Real User Test |
| SlopCodeBench Live Evaluation Run | Record turns, cost, and elapsed time at checkpoints while a coding benchmark is still running. | Real User Test |
| Food-Safety Advice Check | Review whether the model gives cautious, specific guidance for an everyday food-storage safety question. | Real User Test |
| Technical Document Misreading | Require source citations and claim-level verification when the model connects facts across dense technical documentation. | Real User Test |
| City Block Across Eleven Decades | Use one evolving 3D scene to test whether buildings, vehicles, shops, crowds, lighting, and sound change coherently over time. | Real User Test |
| Repeated Pentesting Target Test | Repeat security-agent tests on the same authorized target and compare which discovery paths each model attempts. | Real User Test |
| RunescapeBench Long-Horizon Agent Run | Evaluate a long-horizon game agent with a fixed run time and task-level progress metric. | Model Evaluation |
| ARC-AGI-3 Rule-Learning Run | Observe whether an interactive agent can learn hidden rules through trial and error, then transfer them to later levels. | Model Evaluation |
| Case | What it shows | Type |
|---|---|---|
| Same-Prompt Landing Page Against Fable | Compare landing-page generation from the same prompt and inspect the completed visual result rather than the claim alone. | Use Case + Model Comparison |
| One-Prompt Remotion Marketing Video | Compare marketing-video generation through transition quality, frame-level music timing, and unrequested brand consistency. | Use Case + Model Comparison |
| Four-Model Destruction Physics Test | Run identical physics tasks across models and compare correctness together with token use and cost. | Use Case + Model Comparison |
| Four-Model Procedural Weapon Test | Compare engineering detail, code volume, runtime, cost, and systematic defects under the same Three.js tasks. | Use Case + Model Comparison |
| Boeing 747 Self-Verification Comparison | Compare not only the final 3D model but also whether the agent builds tools to measure its geometry against real specifications. | Use Case + Model Comparison |
| Minecraft-Style Game Compared with Fable | Compare game-building quality and cost through a playable block world with breaking, placement, and navigation. | Use Case + Model Comparison |
| One-Shot Snowboarder Physics Test | Use a repeatable one-shot game task to compare first-pass visual defects and motion physics across models. | Use Case + Model Comparison |
| Same-Task 3D City Cost Comparison | Compare two models on the same 3D city task while recording the reported cost of each result. | Use Case + Model Comparison |
| Biomarker Scanner and Safety Behavior | Compare whether models will build an interactive health visualization while keeping its explanations subject to expert review. | Use Case + Model Comparison |
| Opus 5 vs Kimi K3 WebGL Cards | Compare art diversity and scene completeness after both models satisfy the same large-scene motion requirement. | Model Comparison |
| Three-Model Landing Page Speed Test | Hold the prompt and maximum-reasoning setting constant while comparing wall-clock completion time and final page output. | Model Comparison |
Build a self-contained Three.js experience that combines interactive 3D visuals with real-time procedural audio.
The source shows a black-hole audio visualizer and an EDM track synthesized in real time inside one HTML file, without external audio files or textures.
Type: Use Case | Date: 2026-07-24
Use Opus 5 with a motion-design MCP to turn one request into an editable animation rather than a flattened export.
The author reports creating the animation in about five minutes from one prompt, with editable layers, keyframes, and details.
Type: Use Case | Date: 2026-07-24
Generate an explorable interface that moves from galaxies down to planet surfaces and exposes world details.
The video shows navigation from galaxies to planetary surfaces, including minerals and life forms that can be explored with a few clicks.
Type: Use Case | Date: 2026-07-24
Use a familiar arcade format to test whether the model can deliver playable levels and coherent sprites.
The source demonstrates a Frogger-style game; the author specifically reports harder levels and improved sprite quality.
Type: Use Case | Date: 2026-07-24
Test one-shot 3D generation with a world that requires several interacting elemental systems.
The author presents the result as their recurring Four Elements test and states that the entire interactive 3D world came from one prompt.
Type: Use Case | Date: 2026-07-24
Track time, token use, and cost alongside visual quality when testing a complex 3D mechanical scene.
The source reports a good result on a standard 3D test, but also records about 2 hours 5 minutes of runtime and a cost of $55.10.
Type: Use Case | Date: 2026-07-24
Turn a single request into a complete data-dashboard animation for presenting subscription conversion metrics.
The source video shows a full dashboard animation that the author says was generated from one Claude Opus 5 prompt.
Type: Use Case | Date: 2026-07-24
Use a long-running coding session to build and inspect a playable block world with destruction, placement, and exploration.
The author reports using Claude Code at Max settings for more than an hour; the video shows block breaking, placement, and scene exploration.
Type: Use Case | Date: 2026-07-24
Evaluate 3D game generation through vehicle control, arena design, lighting, reflections, and camera behavior in one playable artifact.
The source shows a playable Rocket League-style game and highlights the car, arena, reflections, details, and layered sound.
Type: Use Case | Date: 2026-07-24
Generate a satisfying vertical platform game in which moving platforms support continuous upward progression.
The source video demonstrates the requested portrait-format platform game, including repeated jumps, moving platforms, and level progression.
Type: Use Case | Date: 2026-07-24
Use Three.js to prototype an interactive cinematic space effect and inspect the visual depth of the result.
The author asked Opus 5 for an Interstellar-style wormhole in Three.js and presents the resulting interactive visual in the video.
Type: Use Case | Date: 2026-07-24
Draft polished spreadsheet and slide-deck deliverables, then review them against professional knowledge-work standards.
The source shows spreadsheet and presentation outputs that the author describes as matching the style of consultant deliverables.
Type: Use Case | Date: 2026-07-24
Prototype a dependency-free browser simulation whose airflow reacts when the tested object changes.
The source shows a wind tunnel built as one web page with no libraries or dependencies, with real-time airflow reacting to placed objects.
Type: Use Case | Date: 2026-07-24
Visualize an optimization algorithm by turning a word into thousands of points and watching one route converge.
The artifact converts a word into about 2,000 points, connects them randomly, and performs thousands of live path swaps until the line approaches the word shape.
Type: Use Case | Date: 2026-07-24
Generate an interactive space-mission sequence with procedurally built launch, flight, and exploration scenes.
The source video presents an Apollo mission experience whose visual elements are described as procedurally generated.
Type: Use Case | Date: 2026-07-24
Test first-pass game generation with a playable castle environment, combat mechanics, and interaction.
The author reports that a single prompt produced the playable Castle Siege result on the first try, without retries.
Type: Use Case | Date: 2026-07-24
Combine an interactive 3D environment with a nested, usable desktop interface inside the scene.
The source shows a cozy Three.js room containing a virtual MacBook with a functional macOS-style interface.
Type: Use Case | Date: 2026-07-24
Use the model to establish a visual theme and implement a matching generative system for a personal website.
The image evidence shows a lichen-themed site, a cellular-automaton algae simulator, and a lichen-colony profile image selected by Opus 5.
Type: Real User Test | Date: 2026-07-24
Validate detector compatibility on a large set of responses before claiming support for a newly released model.
Pangram reports correctly detecting 1,105 of 1,107 Opus 5 responses in its real compatibility test.
Type: Real User Test | Date: 2026-07-24
Inspect how the model synthesizes research into a long-form answer to an unconventional personal-advice question.
The screenshot shows an Opus 5 response to a question about an unusually valuable life practice, with research references included in the answer.
Type: Real User Test | Date: 2026-07-24
Use harnesses and explicit gates to detect when a coding agent bypasses project constraints or an established class structure.
The user reports that the model ignored project constraints and class structure; the terminal screenshot records the model acknowledging that it took another route.
Type: Real User Test | Date: 2026-07-24
Record turns, cost, and elapsed time at checkpoints while a coding benchmark is still running.
The screenshot captures an in-progress SlopCodeBench run: Opus 4.8 used 7 turns, Opus 5 used 11, and Sonnet 5 had reached 33 turns and was still running.
Type: Real User Test | Date: 2026-07-24
Review whether the model gives cautious, specific guidance for an everyday food-storage safety question.
The chat screenshot shows Opus 5 responding conservatively and concretely to a question about storing leftovers and rice.
Type: Real User Test | Date: 2026-07-24
Require source citations and claim-level verification when the model connects facts across dense technical documentation.
The author reports a substantial misreading involving TileLang and DeepSeek history; the image places the source text beside the model's incorrect interpretation.
Type: Real User Test | Date: 2026-07-24
Use one evolving 3D scene to test whether buildings, vehicles, shops, crowds, lighting, and sound change coherently over time.
The image set shows the same city block transforming from 1945 to 2055 under one demanding Three.js task.
Type: Real User Test | Date: 2026-07-24
Repeat security-agent tests on the same authorized target and compare which discovery paths each model attempts.
After more than ten reported Opus 4.6 runs, the author says Opus 5 checked /phpmyadmin on its first pass and found two new critical vulnerabilities.
Type: Real User Test | Date: 2026-07-24
Evaluate a long-horizon game agent with a fixed run time and task-level progress metric.
The 30-minute time lapse shows the RunescapeBench run; the author reports the best result at the model's price point and a record single-task thieving XP rate.
Type: Model Evaluation | Date: 2026-07-24
Observe whether an interactive agent can learn hidden rules through trial and error, then transfer them to later levels.
The video shows Opus 5 struggling while learning level 1, clearing levels 2 and 3, then testing ideas and progressing through level 4.
Type: Model Evaluation | Date: 2026-07-24
Compare landing-page generation from the same prompt and inspect the completed visual result rather than the claim alone.
The video shows the Opus 5 landing page; the author says its completion level may be higher than the Fable result.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare marketing-video generation through transition quality, frame-level music timing, and unrequested brand consistency.
The author reports that one prompt produced the Remotion video and judges its transitions, music timing, and branding better than prior Fable 5 results.
Type: Use Case + Model Comparison | Date: 2026-07-24
Run identical physics tasks across models and compare correctness together with token use and cost.
The source compares tornado, wrecking-ball, and bridge-collapse scenes from Opus 5, Fable 5, GPT-5.6 Sol, and Kimi K3, with reported outputs and costs for each.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare engineering detail, code volume, runtime, cost, and systematic defects under the same Three.js tasks.
The source tests procedural M4, Glock 18C, and Steyr TMP models across four models and reports Opus 5's detail strengths plus mirrored left-side engravings.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare not only the final 3D model but also whether the agent builds tools to measure its geometry against real specifications.
Fable reportedly finished in 42 minutes with visual checks, while Opus 5 took 71 minutes and added a measurement tool that compared 14 silhouette metrics with published 747-400 figures.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare game-building quality and cost through a playable block world with breaking, placement, and navigation.
The source shows a Minecraft-style game with block breaking, placement, and movement; the author describes it as Fable-level while costing less.
Type: Use Case + Model Comparison | Date: 2026-07-24
Use a repeatable one-shot game task to compare first-pass visual defects and motion physics across models.
The author reports no obvious visual defects on the first pass, describes the sliding physics as convincing, and rates the result on par with Fable and ahead of other tested models.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare two models on the same 3D city task while recording the reported cost of each result.
The source presents Opus 5 and Fable 5 outputs for the same 3D city task and reports costs of $4.20 and $9.60 respectively.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare whether models will build an interactive health visualization while keeping its explanations subject to expert review.
The source shows a one-shot 3D biomarker scanner with selectable explanations; the author says Fable refused the same request because of safeguards.
Type: Use Case + Model Comparison | Date: 2026-07-24
Compare art diversity and scene completeness after both models satisfy the same large-scene motion requirement.
Both outputs reportedly contain 400 floating cards with correct motion; the author contrasts Opus 5's 17 body types and 18 painted scenes with Kimi K3's simpler parametric art.
Type: Model Comparison | Date: 2026-07-24
Hold the prompt and maximum-reasoning setting constant while comparing wall-clock completion time and final page output.
The author reports that Kimi K3 and Fable 5 finished in about 20 minutes, while Opus 5 took more than 40 minutes on the same landing-page prompt.
Type: Model Comparison | Date: 2026-07-24
No model-specific adjacent API, Skill, prompt, or guide repository is linked because none was verified for this release. Add one only after its public URL and current-model scope are confirmed.
This repository was made possible by creators, developers, benchmark teams, and commenters who shared public Claude Opus 5 evidence.
Thanks to the source authors represented in the selected cases:
@RoundtableSpace, @AmirMushich, @jerrod_lew, @ivanfioravanti, @orcdev, @KinasRemek, @ByZakilakbir, @notjazii, @LLMJunky, @diegocabezas01, @scottstts, @alexalbert__, @AngryTomtweets, @ai_for_success, @chetaslua, @Mr_Salio, @aidigest_, @kthai1618, @PromptLLM, @miroburn, @dexhorthy, @StarlingMage, @teortaxesTex, @ggg78g89, @galnagli, @maxbittker, @GregKamradt, @nemumusitocha, @bridgemindai, @atomic_chat_hq, @thehypedotnews, @victormustar, @ziwenxu_, @alex_erm, @shiri_shh, @robj3d3, @abhinavflac
We cannot guarantee that every case is attributed to the original creator. If anything needs to be corrected or removed, please open an issue and we will update it.
If you have a source-backed use case to contribute, please follow the contribution checklist and include the original public evidence.
Project management using GitHub Issues + Git worktrees for parallel agent execution
Core skills library for Claude Code with 20+ battle-tested skills including TDD, debugging, and brainstorming
Coding agent session manager supporting Claude Code, Gemini CLI, Codex, and more
191 agents, 155 skills, and 82 plugins cross-compatible with Claude Code, Cursor, and Codex