Video summary

Kimi K3 vs Fable 5 vs. GPT 5.6 (Raw Results)

Main summary

Key takeaways

Product Review

Product(s) compared

  • Kimmy K3 (open-source model; Algretto “$39/month plan” mentioned)
  • Fable 5 (frontier model for heavy 3D)
  • GBT 56 / “56 Soul” (another frontier model)

The video is a structured benchmark-style comparison of how well each model generates WebGPU/3JS real-time 3D worlds, scroll-driven websites, and playable games—plus “everyday” tasks like landing pages and brand refreshes—using closely matched prompts.


Key test results (what the reviewer saw)

1) Real-time 3D “Odysseus whirlpool” scene (ocean + whirlpool)

Setup

  • Same core prompt across all models; added complexity (whirlpool scene, ocean spectrum terms, foam/optics, etc.)
  • “/goal” used to force models to complete objectives.

Generation time

  • Soul: ~40 min
  • Fable: ~2h 24m
  • Kimmy: ~3h

Output / UX observations

Soul

  • Pros
    • Graphics look “pretty good,” good lighting
    • Boat movement feels realistic
    • Has controls (whirlpool strength, sea state, time of day)
  • Cons
    • Whirlpools interaction isn’t great: entering often results in “floating in a circle” with limited steering
    • Grabbing/interaction feels non-functional
    • Time-of-day change feels weak (mostly darkens/lights rather than realistic “golden sun”)

Fable

  • Pros
    • Much more like the reviewer’s intended whirlpool feel
    • Strong reflections
    • Characters look more realistic
    • Lots of visual controls (SSR/AO/DOF/etc.)
    • Adds cinematic camera behaviors
  • Cons
    • Some artifacts/glitches
    • Waves can look “funky” or glitchy
    • Boat/rowing behavior can be glitchy
    • Sometimes objects/artifacts appear as dots
  • Time-of-day (“golden hour”) is improved vs Soul (less just exposure-only).

Kimmy

  • Pros
    • Fewer artifacts than Fable
    • Physics/boat movement feels less glitchy
    • Waves look more natural
    • Controls are decent
  • Cons
    • Follow-camera behavior can be “funky”
    • Camera/cinematic modes feel odd
    • Missing some Odyssey passage detail
    • Whirlpool interaction is closer than Soul but still not exactly as pictured
    • Water quality: reflections can be weaker than Fable

Verdict for this test

  • Fable: best overall fidelity and closest whirlpool behavior (but glitchier)
  • Kimmy: best “less glitchy / closer-than-Soul” option
  • Soul: visually solid, but interaction/game feel is weakest

Cost notes (whirlpool build)

  • Kimmy K3: 187M input / 515K output; $66 without subscription, $39/month plan mentioned
  • Fable 5: $263 total (167M input / 423K output; similar workload/time)
  • Soul (56): $23 total

Reviewer conclusion: Kimmy is far cheaper than Fable for similar quality; Soul is cheapest but lower interaction fidelity.


2) Scroll-driven “IMAX camera” website (3D site scrollytelling + viewfinder scene request)

Generation time

  • Soul: ~40 min
  • Fable: ~2.5 hours
  • Kimmy: ~3+ hours

Runtime/technical issues

  • Soul initially errored (“camera could not be initialized” / WebGL mention), but worked after redeploy.

Output / UX observations

Soul

  • Pros
    • Smooth scroll; “taking apart” camera parts works and looks like real 3D camera
    • Teaching-like visualization
  • Cons
    • Parts sometimes “disappear” or behave oddly
    • Viewfinder scene didn’t really land
    • UI/stat overlays appeared weird

Fable

  • Pros
    • Best overall execution: film grain, bloom around highlights, chromatic aberration
    • Strong cinematic feel
    • Lens textures and lighting look most realistic
    • Better “focus” and storytelling execution
    • Good teaching visualization
  • Cons
    • Rotation/control is annoying or unintuitive
    • Some transitions are funky
    • Highlights don’t perfectly match every expected element

Kimmy

  • Pros
    • Better than Soul in realism
    • Good camera texture, film grain, film strip concept, bloom
    • Fewer artifacts than Fable and Soul in this test
  • Cons
    • Lens placement position problem (“big ding”)
    • Focus controls / scene end didn’t work as expected
    • Some box/camera transitions are awkward
    • Camera viewing angle is sometimes weird

Verdict for this test

  • Fable wins on realism and polish (most cinematic, most “crushed” details)
  • Kimmy is surprisingly close and more usable than Soul, but with notable placement/focus issues
  • Soul is less reliable and less accurate to the intended outcome

Cost notes (scrolling website build)

  • Kimmy K3: $48
  • Fable 5: $238
  • Soul: reviewer notes it “should be much lower,” but it’s stated incorrectly once; later context implies Soul is far cheaper than Fable and Kimmy overall in this category.

Reviewer takeaway: Fable is marginally better output at much higher cost; Kimmy offers best value for “almost-Fable” results.


3) Playable WebGPU game: Odysseus’s Escape from the cave (Cyclops + phases)

Generation time

  • Kimmy: ~57 min
  • Fable: ~40+ min
  • Soul: ~35 min

Despite time differences, the reviewer emphasizes they’re still long tasks.

Output / gameplay observations (overall)

The reviewer says the prompt is extremely complex and models struggle with full “playability” and correct interaction logic.

Soul

  • Pros
    • Playable-ish attempt with stake/heat/aim/thrust-ish controls
    • Some sense of phase objectives
  • Cons
    • Interaction is glitchy/unclear (difficulty determining win/lose)
    • Controls don’t always respond as expected
    • Camera/positioning issues
    • Behavior becomes nonsensical

Fable

  • Pros
    • Arrow firing physics seemed better than Soul
    • Some passage from Odyssey appears
  • Cons
    • Cyclops/cave scale/quality issues
    • Missing control prompts
    • Awkward transitions
    • “Playable” experience fails: cannot interact correctly and objective logic doesn’t make sense

Kimmy

  • Pros
    • Cyclops transition/phase moments (at least stake-heating and firing) feel coherent
  • Cons
    • 3D work not good enough
    • Random rams
    • Hand motion / weird controls
    • Unclear objective completion

Verdict

  • Main point: Fable is still the “3D king,” but all models fall short of fully convincing, robust playability for this extreme prompt.
  • Reviewer frames it as a “stumping” prompt; gameplay interaction logic is where they fail.

Cost notes (game build)

  • Kimmy: $24
  • Fable: $111
  • Soul: $12

The reviewer notes the cost advantage persists even if output quality is closer in other tests.


“Everyday tasks” lightning round

4) Award-winning single-page landing site (3 distinct concepts)

Prompt emphasis

  • Creative freedom, plus “anti-slop” guardrails:
    • No purple/blue gradients
    • No glass morphism
    • No stock hero
  • Reviewer notes the first concept had a “cream background” resembling a known “GPT-56 Soul giveaway” (suggesting model tendencies).

Result ranking

  1. Fable 5 (best overall concept/design)
  2. Soul (56)
  3. Kimmy K3

Reviewer notes Fable’s static site concept quality is better here, but models can produce repetitive or nonsensical filler (e.g., “superolous writing” / astronomy-style lines).

Cost notes (landing page test)

  • Fable 5: $41 (fast: ~18 min)
  • Kimmy K3: $2.75 (cheapest; ~23 min)
  • Soul: $3.22 (still cheap; ~14 min)

Value takeaway

  • Fable wins quality, but Kimmy/Soul are dramatically cheaper and “good enough” for many use-cases.

5) Reviving a dead brand: RadioShack brand revitalization (HTML deck + brand identity + mockups)

Output components required

  • Word mark (SVG), color palette, typography, tagline
  • Brand story/tone
  • Pitch deck slides
  • Image generation for mockups

Pros/cons observed

Kimmy vs others (overall)

  • Reviewer says results are close when tasks aren’t as extreme.
  • Fable appears stronger for conceptual deck execution and polish in some areas.
  • Occasional “dumb logo”/mismatch issues and missing/incorrect assets in mockups were flagged.

Ranking

  1. Fable 5
  2. Soul (56)
  3. Kimmy K3

Still, the reviewer emphasizes: for realistic brand tasks, all three can get quite close—differences are often polish and cost.

Cost notes (brand task)

Relative cost examples provided:

  • Kimmy: ~$1.72
  • Fable: ~$11.94
  • Soul: ~$2.82

The reviewer argues Kimmy becomes extremely attractive for “everyday” work that isn’t heavy 3D.


Overall pros & cons (by product)

Kimmy K3

Pros

  • Frequently closer-than-Soul to frontier quality on 3D visuals (especially whirlpool + scroll site)
  • Often fewer artifacts/glitches than Fable
  • Strong value: far cheaper than Fable; sometimes close to Soul
  • Works well as a daily-driver for websites and brand creation

Cons

  • Scroll site: notable lens placement positioning issues; some focus/scene completion problems
  • Whirlpool: camera/follow controls can be “funky,” and Odyssey details can be missing
  • Game test: still fails on key gameplay/interaction robustness

Fable 5

Pros

  • Best cinematic realism and execution (especially scroll-driven IMAX camera)
  • Best overall heavy 3D results (reviewer calls it “king” of complex long-running 3D)
  • Tends to produce the most impressive polished visuals and control-rich experiences

Cons

  • Very high cost (often described as ~4x or more than Kimmy in similar workloads)
  • More artifacts/glitches in some real-time 3D tasks
  • Still fails to deliver fully robust playability in the extreme game prompt

Soul / GBT 56

Pros

  • Often fastest to complete (e.g., whirlpool build ~40 min)
  • Very cheap compared with Fable; can be competitive on simpler tasks
  • Some 3D behaviors look solid; learning/teaching visualization is a highlight

Cons

  • Interaction/gameplay is weakest in the whirlpool test (e.g., stuck/floating)
  • Scroll site: initial camera initialization failure; intended end scene sometimes doesn’t show
  • Generally less polished than Fable; less reliable for realism/controls

Comparisons emphasized by the reviewer

  • Core theme: quality vs cost tradeoff

    • Fable: best quality for heavy 3D, but “insane” cost
    • Kimmy: surprisingly close output for far less cost; best on “everyday” work
    • Soul: cheapest; sometimes “good enough,” but less consistent and weaker on interactions/realism
  • Workflow strategy

    • Use Fable for very complex, long-running heavy 3D
    • Use Soul for mid-complexity tasks (often cheaper)
    • Use Kimmy as the default daily driver for most work

Unique points mentioned (consolidated list)

  1. Kimmy K3 benchmarks claim impressive performance.
  2. Testing used same prompts across models; multiple build types (3D world, scroll site, playable game).
  3. Confusion about switching Kimmy to “max thinking effort” requiring a config-file change.
  4. Prompt complexity engineered via an external agent (Claude) with 3D/WebGPU terms.
  5. Use of /goal to force completion; Claude Code goal length limits discussed (4,000 chars vs 11,000 prompt).
  6. Deployment to Vercel for live links.
  7. Notes about localhost vs live-host observations (deployment step recorded; performance behavior not deeply emphasized).
  8. Whirlpool interaction differences:
    • Soul: “stuck floating”
    • Fable: more intended spin/entry
    • Kimmy: closer but still imperfect
  9. Controls included whirlpool strength, sea state, time of day, camera modes, and rendering toggles (SSR/AO/DOF/camera shake).
  10. Time-of-day realism differences: - Soul: mostly darkens/lights - Fable: can create “golden hour”
  11. Artifact/glitch discussions: - dots - weird wave artifacts - blotchy discoloration - glitchy rowing
  12. Fable’s reflections and cinematic effects (grain, bloom, chromatic aberration) repeatedly praised.
  13. Kimmy’s film grain and bloom praised, but lens placement and focus behavior flagged.
  14. Game prompt described as “stumping”; gameplay interaction logic often fails.
  15. Cost breakdowns presented per build, including Kimmy subscription vs non-subscription.
  16. Landing page test: “cream background” noted as a likely 56/Soul tendency.
  17. Landing page cost: Kimmy cheapest; Fable still highest quality.
  18. Brand refresh test requirements: word mark SVG, brand assets, deck slides, and mockups via image generation.
  19. Brand refresh ranking: Fable best, Soul second, Kimmy third—yet Kimmy is very cheap for comparable output.
  20. Final recommendation: use open-source Kimmy for most tasks; reserve frontier/high-cost models for the most complex 3D.

Speaker perspectives

  • Single primary reviewer dominates all observations and rankings (no distinct multi-speaker viewpoints clearly separated in subtitles).

Concise verdict / recommendation

  • Need the best cinematic/advanced 3D (especially scroll-driven 3D storytelling)? Choose Fable 5, but expect much higher cost.
  • Want near-frontier quality at a fraction of the price and best value for everyday creative work? Choose Kimmy K3 (the reviewer’s main daily-driver recommendation).
  • Want the cheapest option and can tolerate less consistent interaction/finish? Soul (56) is a valid budget alternative, but it often underperforms on interaction realism and polish.

Original video