1ST PLACERANK 1
v0 by Vercel
Three rounds straight — spec-faithful layout, cleanest markup.
§ 01 · INTRODUCTION
Stop paying frontier prices for trivial work.
Soupy Together routes your AI coding tasks to the tool best suited for each one. An in-house layer that absorbs routine work before any external API call is in active development — until it ships, we route everything to the cheapest capable partner. You pay $29/month, plus only what you actually use.
§ POWERED BY SACE — A PUBLISHED COGNITIVE ARCHITECTURE FOR SIGNIFICANCE-AWARE AI ORCHESTRATION
Soupy Lab LLC · Glen St. Mary, Florida · April 2026 · v1.0
§ 02 · BUILD-OFF
The Build-Off, by category.
A rotating, public competition. Each category is one skill — frontend fidelity, backend correctness, refactor reliability, and so on. Every integrated AI coding tool runs the same task. We publish the podium. Brands earn a badge they can show. You pick which tool you want for each kind of work, and Soupy routes there when the job calls for it.
§ SAMPLE DATA · SOUPY ISN'T A CONTESTANT YET
These podiums are plausible estimates, shown to demonstrate the rubric — not yet verified live runs. Soupy Together is the router, not a ranked tool: it has no in-house engine competing here yet, so it appears only as an honorable mention describing where it would route your work. Once our in-house tier is live and a verified round is published, the rankings here become real.
§ ROUND 003 · SAMPLE · 2026-05-01
Frontend & visual fidelity
Building a polished landing page from a brief. Pixel-accurate UI, layout, responsive design.
YOUR PICK FOR THIS CATEGORY
— not chosen —
saved locally · used as a routing hint on /demo
2ND PLACERANK 2
Lovable Pro
Closed the gap on R3. Strong design system, fastest edit loop.
3RD PLACERANK 3
Bolt
Quick first preview, regressed slightly on responsive breakpoints.
HONORABLE MENTIONS
- Soupy TogetherDaily-driver routing — picks v0 when fidelity is the goal.
§ YOUR ROUTING PREFERENCES
Your picks travel with you to the demo. The router uses them as hints — budget and availability still get the final word.
- Frontend & visual fidelity—
- Backend correctness—
- Refactor reliability—
- Data & SQL—
- Honesty under uncertainty—
- Cost per output—
§ HOW TO READ IT
Soupy Together is the daily driver, not the top scorer. We don't need to win — we need to know who does. When your job lands in a category where another tool ranks first, that's where we route it. You only pay for that call.
§ FOR PARTNERS
Rank 1st, 2nd, or 3rd in any category and you get an embeddable SVG badge for your site. No tool can pay for placement. The rubric is public; the receipts are public.
§ 03 · THE PROBLEM
Every AI coding tool hits a wall.
CAPABILITY MISMATCH
Lovable is great for frontends. Cursor is great for refactors. Claude Code is great for backends. None of them know when to hand off. You hit a wall, give up, or pay credits to fake what they can't do.
FRONTIER OVERKILL
Renaming a button shouldn't cost the same as architecting a service. But every tool routes everything to its most expensive model. You burn $200/month in credits on $5 worth of actual reasoning.
OPAQUE PRICING
Credits don't map to dollars. You can't compare across vendors. You can't budget. You find out you're broke when the cap hits, not when the work was done.
§ 04 · ARCHITECTURE
Anatomy over monolith.
Soupy Together is built on SACE — the Significance-Aware Cognitive Engine, a published cognitive architecture that mirrors how biological nervous systems handle attention. Most processing is meant to happen in small specialized models near the sensor stream, with expensive frontier models dormant by default and woken only on interrupt — when consensus across modalities indicates a significant event. The design target is a system that handles 60% of requests in-house for cents, escalates 30% to specialized partners, composes 9% across multiple tools, and only invokes a frontier model on the rare 1% that genuinely require it. Today the routing and partner layers are live; the in-house Tier 0 model is still in development, so that 60% is currently routed to the cheapest capable partner rather than absorbed locally.
TIER 0
Local Absorption
—
TIER 1
Single Partner
30%
TIER 2
Multi-Partner
9%
TIER 3
Cortex Escalation
1%
§ LIVE ENGINE · DETERMINISTIC SIMULATION
User asks for a real feature. Consensus rises across modalities.
OBSERVED
0
INTERRUPTS
0
PROMOTION
0.0%
CURRENT TIER
TIER 0
CHROMATIC VECTOR · COMPOSITE 0.000 · THRESHOLD 0.45
Red
0.00
Blue
0.00
Yellow
0.00
Green
0.00
Orange
0.00
Local absorption — no external call
TIER ESCALATION · COMPOSITE GATES PROMOTION DEPTH
§ AWAITING FIRST CROSSING
TIER 0
0.25
TIER 1
0.50
TIER 2
0.75
TIER 3
1.00
TIER 0
Local absorption — no external call
ENTERED
0
PROMOTED
0
TIER 1
Routed to a single specialist partner
ENTERED
0
PROMOTED
0
TIER 2
Composed across multiple partners
ENTERED
0
PROMOTED
0
TIER 3
Frontier cortex escalation
ENTERED
0
PROMOTED
0
§ COMPOSITE CROSSES A BAND → CORTEX TARGETS THAT TIER · PROMOTION FIRES ONLY WHEN INSTANT COMPOSITE ≥ 0.45
RING BUFFER · TTL 4s
§ AWAITING OBSERVATIONS
PROMOTED MEMORIES
§ NO PROMOTIONS YET — MOST OBSERVATIONS EXPIRE UNREAD
Runs entirely in your browser. Same cortex synthesis logic published in the SACE paper. No LLM call, no telemetry — refresh to reset.
§ 05 · PRICING
What it actually costs to build with AI.
Updated weekly. Source links below.
| Tool | Monthly | Notes |
|---|---|---|
| ChatGPT Plus | $20 | Conversational only — no orchestration |
| Cursor | $20 | IDE assistant — requires existing dev skill |
| v0 by Vercel | $20 | Frontend only |
| Lovable Pro | $25 | Frontend + Supabase, credit-capped |
| Replit Core | $25 | Plus separate agent credits |
| Bolt | $20 | Frontend, credit-capped |
| Claude Pro | $20 | Conversational, no orchestration |
| ChatGPT Pro | $200 | Frontier model access |
| Claude Max | $200 | Frontier model access |
| Soupy Together | $29 | Routes across all of the above. Pay only for what gets called. |
Prices reflect publicly listed subscription rates as of April 2026. We do not accept payment for placement or rankings. Methodology and receipts: build-off ledger →
Soupy Together aims to be cheaper because most of what you ask AI to do — rename a variable, add a route, write a query, scaffold a component — doesn't need a $200/month frontier model. Today we route the small stuff to the cheapest capable tool; an in-house tier to absorb it for near-zero cost is in development. Either way, we only reach for premium tools when you actually need premium output — so you stop paying frontier prices for trivial work.
§ 06 · INTEGRATIONS
Built to plug into the routing & observability stack.
Soupy is the significance layer above the wire. If you run an LLM gateway, router, or observability platform, Soupy's tier classification, capability matrix, and signed receipts are designed to feed into yours — not replace them.
FOR LLM GATEWAYS
OpenRouter · Portkey · LiteLLM
Consume Soupy's tier-0/1/2/3 classification as a routing hint. We decide what's trivial vs. frontier-worthy; you handle the wire, fallbacks, and key vaulting. Cleaner separation than re-implementing classification per gateway.
FOR OBSERVABILITY
Helicone · Langfuse · Arize
Every Soupy call emits a signed receipt with task, tier, partner, and cost. Ingest the receipt stream as a first-class span source — tier-tagged spend, partner attribution, and build-off provenance without instrumenting each tool.
FOR ROUTERS
Martian · Not Diamond · RouteLLM
Model-level routing is solved. Task-level routing across tools (Cursor vs. Lovable vs. Claude Code) isn't. Soupy's capability matrix is the missing input — license it, or compose with it.
§ INTEGRATION PROGRAM · Q3 2026
Receipt schema, capability matrix export, and webhook spec are in private review. If you're building a gateway, router, or observability layer and want Soupy's signal feeding yours, get on the integration list — separate from end-user early access.
§ 07 · TECHNICAL DISCLOSURE
Read the architecture.
Soupy Together is the commercial implementation of work timestamped and published by Soupy Lab. The papers are open technical disclosures.
WHITEPAPER · v1.0 · APRIL 2026
SACE: A Significance-Aware Cognitive Architecture with Retroactive Memory Formation
The foundational architecture: anatomy over monolith, retroactive labeling, interrupt-driven cognition, polychromatic decay, and honest null states.
Request preprint →WHITEPAPER · v1.0 · APRIL 2026
Soupy Together: A Cost-Disciplined Multi-Agent Orchestration Platform
The product paper: four-tier execution, capability matrix methodology, transparent pricing, and additive partner economics.
Request preprint →WHITEPAPER · v1.0 · APRIL 2026
SACE Applied: A Significance-Aware Architecture for Multi-Agent Software Development
The bridge paper: how SACE's foundational primitives translate from conversational memory to development orchestration.
Request preprint →§ 08 · EARLY ACCESS
Built by a guy who pays for all of them.
Soupy Together is built by John "Soupy" Beene — marine construction superintendent, Army veteran, solo founder of Soupy Lab. I pay for Claude Pro, ChatGPT Plus, Lovable, Cursor, and Replit every month. I got tired of the credit math and built the tool I wanted to exist. If that sounds like a problem you've had, get on the early access list.
§ NO SPAM · NO INVESTOR-SPEAK · NO LAUNCH HYPE
§ YOU GET ONE EMAIL WHEN THE BUILD-OFF DROPS · ONE EMAIL WHEN ACCESS OPENS
§ THAT'S IT