GPT-6.1 Sol vs Claude Sonnet 5.5 – This Was NOT Even CLOSE!

Product Reviewed

Two AI models were compared head-to-head across multiple tasks:

• Anthropic Claude Sonnet 5.5 (“Sonnet 55”)
• OpenAI GPT-6.1 (named “Soul” / “GPT 6.1 Soul”) (“Soul”)

The video frames this as a capability/product comparison in “real-world” workflows, including:

• 3D/game building replication
• robot software + branding + marketing deliverables
• physical robot/RC control
• music and video production

---

Key Features & Capabilities Tested (and What Happened)

1) Image → Blender/GDAU Game Replication (Demolition Derby)

Hard requirements

• soft-body crash deformation
• exact-ish car models (Ford Crown Victoria)
• 3 camera views
• sound effects
• suspension/engine visibility
• realistic part fall-off behavior

GPT-6.1 Soul (≈ 1h 40m on max effort)

Strong points

• good crash and sound-effects behavior
• muffler falls off and changes loudness
• visible engine, suspension, and tire puncture
• moving steering wheel
• parts falling off

Weak points

• not truly pixel/shape replication
• presenter notes mismatched modeling/physics fidelity versus an “exact match” expectation

Claude Sonnet 5.5 (max effort)

Strong points

• more faithful car curvature/details
• cinematic camera follow behavior
• broken glass, debris
• suspension and engine model visible

Weak points / bugs

• detached parts may have collider issues (e.g., a door falling off but still blocking collision)
• some “weird sound” interactions

Overall (for this test): The presenter felt Sonnet 55 was closer to the reference.

---

2) Real-World “Co-Founder” Task: Robot Chatbot Redesign + Branding + Website + Ad + Working Software

Product context: “Perry” A desktop robot voice chat buddy (Raspberry Pi, screen, head movement, speaker/mic).

Provided to the models

• outdated software + instructions
• 3D assets (for Keyshot-style renders)
• overhaul requirements:
  • UI/branding/website
  • 15s motion ad
  • voice integration (OpenAI real-time voice API)
  • deployable folder + instructions

Claude Sonnet 5.5 Outcome

• Website quality: very high
  • high-end look
  • interactive UI demos
  • brand guidelines
  • multiple robot renders/finishes
  • logo embossing created even though not in source assets
• Marketing video
  • generated and formatted for 9:16 and 16:9
• Software demo in live testing
  • head movement + visual listening states
  • responsive voice
  • volume adjustment via voice
  • timer support
  • moods/states

Minor concerns

• could miss small realism details (e.g., screw visibility)
• some brief UI “bugging out”
• back-side less polished due to manufacturability constraints

GPT-6.1 Soul Outcome

• Website
  • decent and includes interactivity/live preview/captions
  • presenter notes render/UI offsets and “jank”
  • frontend resembles Claude branding more than expected
• Software
  • worked, but presenter reports it did not support some voice controls well (notably voice volume adjustment)
  • demo can adjust “vibe,” but not select a specific named voice (e.g., “Cedar”)
• Presenter conclusion: GPT-6.1 was functional but less refined in sales/marketing polish.

Overall (for this test): The presenter strongly preferred Claude Sonnet 5.5:

• “Perry” branding/website/UI/marketing were far more launch-ready
• GPT-6.1’s marketing deliverables were weaker (even if the software “worked”)

---

3) Physical “Parallel Parking” RC Car via Camera + Arduino Frequency Clone (Top-Down)

Tasks

• control an RC car using a cloned Arduino RF radio frequency
• perform careful parking with no proportional control (on/off only)
• handle low traction (wheels spin; car veers)
• output a 20–30s high-quality methodology motion graphic after the run

GPT-6.1 Soul

Strong points

• methodical; didn’t “yeet” the car
• camera-guided pulse timing and pose-estimation logic
• produced an excellent narrated 29s infographic explaining the approach

Weak points

• didn’t verify whether simultaneous turn+move commands worked (instruction limitation)
• parking ended incomplete before time cutoff

Claude Sonnet 5.5

Strong points

• methodical
• tried “two actions simultaneously” despite being unverified (viewed as a positive divergence)
• asked about handling a gap-size issue and then continued using bump-parking guidance
• produced a motion graphic without voiceover, with more visual animation

Big negative

• presenter claims Sonnet flagged the request as a potential “reasoning extraction/distillation attack”, interrupting with safety framing and disrupting the workflow

Overall (for this test): No clear winner, but Sonnet is faulted for the safety-interruption behavior.

---

4) 50s Rap Album Promo: Mix/Master Stems + Autotune + Video Editing/Motion Design

Inputs

• rap stems (instrumentals + recorded iPhone vocal)
• project constraints (GarageBand-style)
• b-roll clips + album cover
• prompt to produce a polished promo video + mastered track

GPT-6.1 Soul

• audio remix/mix was worse than expected
  • voice sounded “tiny”
  • lows not great
  • closer to the source than a true remix
• video was decent but less impressive visually than Sonnet
• some overlays/transitions were okay

Claude Sonnet 5.5

• significantly better across:
  • instrumental changes/additional sound design
  • doubled vocals with audible undertone (as described)
  • higher overall video quality and better color consistency
  • more polished transitions and effects

Overall (for this test): Sonnet 5.5 decisively wins.

---

Comparisons Made

• Direct, repeated comparisons across the same four tasks
• The presenter emphasizes:
  • Soul “barely outperforms” or is less ambitious
  • Sonnet “mops the floor” in objective replication and creative production tasks

---

Pros (as Demonstrated)

Claude Sonnet 5.5

• more faithful 3D/game replication (better curvature/details; closer to reference)
• major advantage in product marketing deliverables
  • website, UI demo interactivity, branding system, renders, ad video
• stronger creative output in the rap promo
  • better mix/master and more sophisticated video editing/effects
• physical control: methodical performance; even attempted simultaneous steering+movement

---

Cons / Negatives (as Demonstrated)

Claude Sonnet 5.5

• physical-task safety behavior: interruption for suspected reasoning extraction/distillation attack
• derby game: collision/mesh handling issues on detached parts
• minor website/visual nitpicks (e.g., screw detail not present; brief UI bug; back-side less polished)

GPT-6.1 “Soul”

• often less ambitious creatively (especially music/video mixing)
• robot “Perry” branding/web output: more render/UI jank and less refined marketing polish
• some voice-interaction limitations in the presenter’s experience (e.g., volume control via voice)

---

Overall User Experience / Workflow Feel (from the Presenter)

• Both models can complete complex tasks.
• Claude feels more launch-ready and more willing to produce polished assets end-to-end (website + ad + UI + integration).
• GPT-6.1 is described as a workhorse that gets things working, but sometimes with less refinement and more minimalism in creative tasks.
• Physical-task behavior differed:
  • GPT-6.1 provided clear methodology without the claimed safety disruption
  • Claude produced a good infographic but (per the presenter) triggered safety-style interruptions when details were requested

---

Numerical Scores / Ratings

• No explicit numeric ratings (no stars/percent scores).
• Timing details mentioned:
  • GPT-6.1 Soul: often ~1h 40m for major tasks; music/video completed in ~54 minutes
  • Claude Sonnet 5.5: timing not consistently remembered; “max effort/ultra code”
  • Physical parking: both were around ~1 hour (presenter: “a little over an hour” each)

---

Unique Points Mentioned (Consolidated)

1. Testing approach: 4 real-world/objective-ish tasks; mostly “max” reasoning effort.
2. Derby constraints: soft-body deformation, exact-ish car modeling, 3 camera views, SFX, suspension/engine visibility, part fall-off.
3. GPT derby: strong SFX and engine/suspension visibility, but not faithful replication (admitted shortcuts/prototype physics approach).
4. Sonnet derby: more faithful curvature/cinematics, but collider issues on detached parts.
5. Robot “Perry” requirements: outdated Raspberry Pi chatbot overhaul + UI/branding + Keyshot-style renders + 15s motion ad + deployable folder/instructions.
6. Sonnet’s “Perry” advantage: logo embossing, interactive website, multiple finishes, strong sales UX.
7. Sonnet’s software demo: voice volume change and timers; GPT limited voice controls (as experienced).
8. Parking challenge: RC car controlled via cloned Arduino RF; on/off control only; low traction.
9. GPT parking: camera-guided pulse strategy + strong narrated 29s infographic; incomplete before cutoff.
10. Sonnet parking: tried simultaneous commands; infographic more visual but no voiceover; safety interruption about “reasoning extraction/distillation attack.”
11. Rap promo required mixing/mastering stems, autotune alignment, video editing to match narrative, and motion design beyond basic assembly.
12. GPT music/video: worse audio mix (tiny voice, weaker lows) and less impressive video.
13. Sonnet music/video: better remixed instrumental, added sound design, doubled vocal undertone, higher-quality visuals, better transitions/effects.
14. Final: Sonnet 5.5 “mopped the floor”; GPT-6.1 is capable but less refined/minimalistic; choose Sonnet for launching.

---

Speaker Views

• Only one primary presenter is identified (“I’m Carrie.” / “the presenter”).
• No other distinct reviewers are treated as separate sources of evaluation.

---

Concise Verdict / Recommendation

Verdict: Choose Anthropic Claude Sonnet 5.5 over OpenAI GPT-6.1 “Soul” for these end-to-end product/creative workflows.

In the derby replication—and especially in the robot product marketing + music/video production tests—the presenter portrays Sonnet as far more polished and launch-ready. GPT-6.1 is presented as a capable “gets it working” workhorse, but with weaker refinement and more minimal output in the presented results.