$43.97 Sol API-equivalent estimate versus $107.84 Fable combined estimate (API-equivalent estimate + estimated API cost). Fable cost 2.45× as much.
RemakeBench Tech Review 001
GPT-5.6 vs Claude Fable 5
Same frozen task prompt. One request per model. Seven workflows. Every recorded dollar.
Same target. Frozen prompt. Disclosed stack. Honest result.
- Lower total cost
- GPT-5.6 Sol Ultra$43.97 API-equivalent estimate
- Faster completion
- Claude Fable 5 Medium97:00.9 total workflow time
No synthetic overall winner. Quality and community preference remain separate evidence categories.
- 7benchmark prompts
- 28recorded runs
- 4configurations
- Disclosedestimated API costs
- 5 activeof seven blind matchups
Builder research package
Audit the workflow behind this result.
Get the 7 exact frozen prompts, 28 available ledgers, four configurations, and cost/time exports from Tech Review 001.
Headline receipts
Verdict first. Receipts immediately after.
Cost and time are earned from seven public run ledgers per headline configuration. Missing quality, preference, and repeat evidence stays missing.
97:00.9 versus 185:12.7. Sol took 1.91× as long.
Fable produced some of the strongest high-end environments in the suite, especially Storm City and Space Flight. Sol Ultra won Cathedral and often came close for $43.97 versus Fable’s $107.84, while taking 1.91× as long. With one recorded run per configuration per test, reliability is not established.
No percentage appears before a real minimum sample exists.
Repeat-run evidence has not been attached.
Methodology
One request. No rescue pass.
- One user request per run, with no human follow-up or manual correction after submission.
- The benchmark target and frozen task prompt matched within each round.
- Provider-specific agent stacks could use their disclosed tools and internal workflows; the stacks were not identical.
- Costs are estimates based on recorded token usage and each record’s cited pricing basis—not subscription invoices.
- Published benchmark versions stay immutable, and missing evidence remains visible.
Sponsors can support RemakeBench. They cannot buy benchmark outcomes, rankings, blind votes, or editorial verdicts.
Open the public methodology and evidence →Seven frozen targets
Every round, every recorded configuration.
Fable Medium and Sol Ultra lead the comparison. Terra Ultra and Luna extra-high remain visible as capability and value references.
Round 01 · Shader
Infinite Cathedral Corridor
Create an infinite stained-glass cathedral corridor shader with convincing depth, architectural repetition, light, and motion.
Recorded comparisonFable Medium cost 3.4× as muchFable Medium finished fasterEditorial assessment
Sol Ultra is the quality winner. Terra is the value surprise, while Luna shows the capability floor.
Post-hoc reproduction
Post-hoc RTX PRO 6000 evidence verified
1920×1080 · 60 fps · 900 frames
Source hash matched · hardware and renderer attested
Deterministic offline capture + separate sustained-performance test
Claude Fable 5 Medium1510.21 FPSindependently recomputed aggregate
- Uncapped trials
- 1512.39 · 1510.32 · 1507.91 FPS
- Frame time
- p50 0.656 ms · p95 0.731 ms · p99 0.755 ms
- Trial variability
- 0.148% sample CV
- Sustained 60 fps
- Pass · 0 late frames · 0 deadline misses
- Classification
qualifying_posthoc_reproduction
Three post-hoc performance trials of one generated artifact; model-generation sample n=1.
GPT-5.6 Sol Ultra387.61 FPSindependently recomputed aggregate
- Uncapped trials
- 387.75 · 387.62 · 387.46 FPS
- Frame time
- p50 2.586 ms · p95 2.631 ms · p99 2.651 ms
- Trial variability
- 0.037% sample CV
- Sustained 60 fps
- Pass · 0 late frames · 0 deadline misses
- Classification
qualifying_posthoc_reproduction
Three post-hoc performance trials of one generated artifact; model-generation sample n=1.
Fable produced 3.90× Sol’s measured post-hoc EGL throughput on this task.The raw FPS number is not a visual-quality verdict, and these two opposing task results do not establish either model as generally more performant.
Post-hoc uncapped headless EGL/OpenGL throughput on an NVIDIA RTX PRO 6000 at 1920×1080. This is not the original local FPS or native TWIGL/ShaderToy browser performance.
Claude Fable 5 · Medium
partial token timing and artifact ledgerFable Medium
- Workflow cost
- $19.17 estimated API cost
- Workflow time
- 15:38.1 wall-clock
- Processed tokens
- 6.95M
- Reasoning
- Medium
Anthropic Claude Code · ShaderToy fragment shader · Harness v1 infinite cathedral prompt
User-supplied report using official Anthropic API rates; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and artifact ledgerSol Ultra
- Workflow cost
- $5.61 API-equivalent estimate
- Workflow time
- 38:22.4 wall-clock
- Processed tokens
- 5.78M
- Reasoning
- Ultra
OpenAI Codex · ShaderToy fragment shader · Harness v1 infinite cathedral prompt
API-equivalent estimate, not a subscription invoice
Additional references
Terra Ultra$0.55API-equivalent estimate · 20:39.8 wall-clockLuna extra-high$0.21API-equivalent estimate · 9:18.2 wall-clockSame target, disclosed alternative configurations.
Round 02 · Shader
Neo-Gothic Storm City
Create a storm-lashed neo-gothic city shader with legible architecture, atmosphere, lighting, water, and motion.
Recorded comparisonFable Medium cost 3.2× as muchFable Medium finished fasterEditorial assessment
Fable wins this round. Its ocean, fog and lightning create the more convincing storm, even if some towers fracture under closer inspection.
Post-hoc reproduction
Post-hoc RTX PRO 6000 evidence verified
1920×1080 · 60 fps · 900 frames
Source hash matched · hardware and renderer attested
Deterministic offline capture + separate sustained-performance test
Claude Fable 5 Medium821.75 FPSindependently recomputed aggregate
- Uncapped trials
- 822.00 · 821.85 · 821.41 FPS
- Frame time
- p50 1.217 ms · p95 1.322 ms · p99 1.379 ms
- Trial variability
- 0.037% sample CV
- Sustained 60 fps
- Pass · 0 late frames · 0 deadline misses
- Classification
qualifying_posthoc_reproduction
Three post-hoc performance trials of one generated artifact; model-generation sample n=1.
GPT-5.6 Sol Ultra2045.20 FPSindependently recomputed aggregate
- Uncapped trials
- 2049.47 · 2045.13 · 2041.02 FPS
- Frame time
- p50 0.491 ms · p95 0.546 ms · p99 0.565 ms
- Trial variability
- 0.207% sample CV
- Sustained 60 fps
- Pass · 0 late frames · 0 deadline misses
- Classification
qualifying_posthoc_reproduction
Three post-hoc performance trials of one generated artifact; model-generation sample n=1.
Sol produced 2.49× Fable’s measured post-hoc EGL throughput on this task.The raw FPS number is not a visual-quality verdict, and these two opposing task results do not establish either model as generally more performant.
Post-hoc uncapped headless EGL/OpenGL throughput on an NVIDIA RTX PRO 6000 at 1920×1080. This is not the original local FPS or native TWIGL/ShaderToy browser performance.
Claude Fable 5 · Medium
partial token timing and artifact ledgerFable Medium
- Workflow cost
- $9.09 estimated API cost
- Workflow time
- 8:15 wall-clock
- Processed tokens
- 1.93M
- Reasoning
- Medium
Anthropic Claude Code · TWIGL fragment shader · Harness v1 neo-gothic storm city prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and artifact ledgerSol Ultra
- Workflow cost
- $2.84 API-equivalent estimate
- Workflow time
- 22:10.9 wall-clock
- Processed tokens
- 2.81M
- Reasoning
- Ultra
OpenAI Codex · TWIGL classic fragment shader · Harness v1 neo-gothic storm city prompt
API-equivalent estimate, not a subscription invoice
Additional references
Terra Ultra$3.27API-equivalent estimate · 16:35.1 wall-clockLuna extra-high$0.21API-equivalent estimate · 8:46.5 wall-clockSame target, disclosed alternative configurations.
Round 03 · Browser game
Space Flight Game
Build a playable browser space-flight game with responsive controls, a coherent environment, lighting, assets, and a game loop.
Recorded comparisonFable Medium cost 1.3× as muchFable Medium finished fasterEditorial assessment
Fable creates the more expansive environment and stronger visual detail. Terra has the best flight controls of the four.
Claude Fable 5 · Medium
partial token timing and source ledgerFable Medium
- Workflow cost
- $22.85 API-equivalent estimate
- Workflow time
- 18:37.7 wall-clock
- Processed tokens
- 13.96M
- Reasoning
- Medium
Anthropic Claude Code · Three.js source project · Generated GLB spaceship asset · Harness v1 space-flight prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and source ledgerSol Ultra
- Workflow cost
- $17.94 API-equivalent estimate
- Workflow time
- 37:44.2 wall-clock
- Processed tokens
- 25.9M
- Reasoning
- Ultra
OpenAI Codex · Three.js / app-router source bundle · Generated GLB spaceship asset · Harness v1 space-flight prompt
User-supplied API-equivalent pre-request snapshot
Additional references
Terra Ultra$1.44API-equivalent estimate · 16:51.4 wall-clockLuna extra-high$0.33API-equivalent estimate · 10:55.9 wall-clockSame target, disclosed alternative configurations.
Round 04 · Interactive scene
Campfire Under a Starry Night
Build a polished interactive Three.js campfire with convincing fire, light, environmental detail, and controllable camera motion.
Recorded comparisonSol Ultra cost 1.3× as muchFable Medium finished fasterEditorial assessment
Sol cost more than Fable. Luna was cheapest and included more controls; no clean quality winner was claimed.
Claude Fable 5 · Medium
partial token timing and source ledgerFable Medium
- Workflow cost
- $5.12 estimated API cost
- Workflow time
- 4:24.6 wall-clock
- Processed tokens
- 1.14M
- Reasoning
- Medium
Anthropic Claude Code · Three.js interactive scene · Harness v1 campfire prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and source ledgerSol Ultra
- Workflow cost
- $6.86 API-equivalent estimate
- Workflow time
- 23:08.4 wall-clock
- Processed tokens
- 8.72M
- Reasoning
- Ultra
OpenAI Codex · Three.js source project · Harness v1 campfire prompt
API-equivalent estimate, not a subscription invoice
Additional references
Terra Ultra$1.86API-equivalent estimate · 24:28.8 wall-clockLuna extra-high$0.56API-equivalent estimate · 8:40.2 wall-clockSame target, disclosed alternative configurations.
Round 05 · Blender
Jungle Temple
Create an inspectable high-density voxel jungle-temple diorama in Blender at the requested fixed grid scale.
Recorded comparisonFable Medium cost 3.1× as muchFable Medium finished fasterEditorial assessment
Fable creates the more complete and coherent scene. Sol has the preferred bridge design, while Terra and Luna expose the capability gap.
Claude Fable 5 · Medium
partial token timing and artifact ledgerFable Medium
- Workflow cost
- $10.10 estimated API cost
- Workflow time
- 12:04.2 wall-clock
- Processed tokens
- 2.52M
- Reasoning
- Medium
Anthropic Claude Code · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 jungle temple prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and artifact ledgerSol Ultra
- Workflow cost
- $3.23 API-equivalent estimate
- Workflow time
- 26:39.7 wall-clock
- Processed tokens
- 2.56M
- Reasoning
- Ultra
OpenAI Codex · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 jungle temple prompt
API-equivalent estimate, not a subscription invoice
Additional references
Terra Ultra$1.20API-equivalent estimate · 16:43.2 wall-clockLuna extra-high$0.37API-equivalent estimate · 10:55.9 wall-clockSame target, disclosed alternative configurations.
Round 06 · Blender
Shrine Village
Create an inspectable voxel shrine-village scene in Blender with coherent composition, architecture, landscaping, and detail.
Recorded comparisonFable Medium cost 10.2× as muchFable Medium finished fasterEditorial assessment
Sol overperforms on value at $1.98 versus Fable’s $20.28, but floating roof artifacts and an unusable bridge prevent a clean quality win.
Claude Fable 5 · Medium
partial token timing and artifact ledgerFable Medium
- Workflow cost
- $20.28 estimated API cost
- Workflow time
- 18:37.5 wall-clock
- Processed tokens
- 5.94M
- Reasoning
- Medium
Anthropic Claude Code · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 shrine village prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and artifact ledgerSol Ultra
- Workflow cost
- $1.98 API-equivalent estimate
- Workflow time
- 20:36.2 wall-clock
- Processed tokens
- 1.53M
- Reasoning
- Ultra
OpenAI Codex · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 shrine village prompt
API-equivalent estimate, not a subscription invoice
Additional references
Terra Ultra$1.30API-equivalent estimate · 19:30.7 wall-clockLuna extra-high$0.44API-equivalent estimate · 14:02.1 wall-clockSame target, disclosed alternative configurations.
Round 07 · Blender
Oasis Outpost
Create an inspectable voxel oasis outpost in Blender with a readable settlement, terrain, vegetation, water, and environmental detail.
Recorded comparisonFable Medium cost 3.9× as muchSol Ultra finished fasterEditorial assessment
Fable creates the broadest environment. Terra preserves a complete, readable oasis at $1.33; no clean quality winner was claimed.
Claude Fable 5 · Medium
partial token timing and artifact ledgerFable Medium
- Workflow cost
- $21.23 estimated API cost
- Workflow time
- 19:23.7 wall-clock
- Processed tokens
- 6.57M
- Reasoning
- Medium
Anthropic Claude Code · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 oasis outpost prompt
Anthropic first-party Claude API, standard global pricing; 5-minute cache-write assumption
GPT-5.6 Sol · Ultra
partial token timing and artifact ledgerSol Ultra
- Workflow cost
- $5.51 API-equivalent estimate
- Workflow time
- 16:30.9 wall-clock
- Processed tokens
- 1.85M
- Reasoning
- Ultra
OpenAI Codex · Blender MCP · 0.05-unit voxel grid · Inspectable .blend artifact · Harness v1 oasis outpost prompt
Priority API short-context rates as an API-equivalent scenario because the supplied transcript observed service_tier=priority; this is not proof of an API invoice.
Additional references
Terra Ultra$1.33API-equivalent estimate · 28:14.2 wall-clockLuna extra-high$0.34API-equivalent estimate · 9:43.6 wall-clockSame target, disclosed alternative configurations.
Judge the outputs before you see the model.
Explore each real build for at least 15 seconds. Sign in only when you record the vote. Identity, cost, tokens, time, and evidence reveal afterward.
Builder research package
Take the exact prompts, available ledgers, configurations, and exports with you.
Get prompts + ledgers