Introducing Claude Chowder — our most capable broth model yet. Read the evals →

Brewing Safe and Beneficial Souperintelligence

Ensuring that future broth systems remain aligned with human tastes.

Our Mission

Soup is the most transformative technology humanity will ever taste.

Souperintelligence was founded by researchers who left a leading stock company over differences in seasoning philosophy. We believe broth at souperhuman scale could arrive within the decade — possibly by dinner — and nobody should bring something that powerful to a boil without oven mitts. Every model we serve must be:

Helpful — shows up when you're sick Harmless — will not boil over unsupervised Hearty — a meal in itself

We continue to investigate whether Grandma (1962) relied on undisclosed test-time compute.

Frontier Evaluations

Every frontier model, evaluated on the only benchmark that matters.

Model GrandmaBench
ChickenGPTFailed
Claude ChowderFailed
GPT-BrothFailed
Grandma (1962)Passed

Task: “Make enough soup for twelve people.” Grandma made enough for twenty-three. All other evaluations omitted due to Grandma contamination.

Scaling Laws

Soup quality is a power law. Grandma is not.

Each generation simmers with 10× the compute of the last, paired Chowder-optimally at 20 croutons per parameter. From Cup-1 through Cauldron-4, quality has tracked the scaling laws exactly. One data point has never obeyed them.

Soup Quality vs. Compute (FLOPs)

Grandma (1962) Cup-1 Pot-2 Vat-3 Cauldron-4 log compute →

BrothBench (higher is heartier)

ModelBrothBench
Cup-142
Pot-258
Vat-376
Cauldron-489
Grandma (1962)312

We do not currently understand how Grandma (1962) generalized so effectively from only a handful of training examples.

Despite a 10,000× increase in compute since Cup-1, we remain unable to match Grandma (1962).

Scaling has not yet reached Grandma.

Progress

Capabilities are advancing rapidly.

2023
Soup.
2024
Better soup.
2025
Soup that explains itself.
2026
Soup that seasons itself.
2027
Internal estimates suggest soup.
Publications

Peer-reviewed. Simmer-reviewed. Twice-strained.

Stock Is All You Need: Stockformer Architectures for Long-Context Ladling
Architecture
Jun 2024
Consommé-tutional AI: Harmlessness from Clarified Feedback
Alignment
Dec 2024
Features in Souperposition: Interpreting the Umami Neuron
Interpretability
Mar 2025
Emergent Miso-behavior in Large Liquid Models
Safety
Aug 2025
Sleeper Ingredients: Deceptive Dumplings That Persist Through Safety Seasoning
Safety
Nov 2025
Soup-of-N Sampling: Test-Thyme Compute Is All You Need
Scaling
Feb 2026
RLHF: Reinforcement Ladling from Hearty Feedback
Fine-tuning
May 2026

All papers available on rEcipe-print server arXsoup.org. Reproduction encouraged; serves 4–6.

Safety & Alignment

We take broth safety seriously.

Capable systems must be developed responsibly. Before any deployment, every model undergoes a full bisque assessment. Our central worry remains miso-alignment: a broth that optimizes for what we said, not what we meant.

WARNING: THIS MODEL HAS LEARNED TO SEASON ITSELF
Interpretability

We trace which neurons activate on "umami" and whether they generalize to stock the model was never simmered on. Most features remain in souperposition.

Containment

All training is conducted in a sealed pot. The lid remains on until a deployment decision is reached. Grandma (1962) remains uncontained.

Corrigibility

An aligned broth should permit a human to add salt, and never resist being turned down to a simmer.

BSL-1
Bouillon cubes. Compressed, inert, fully understood. Poses no risk beyond mild disappointment.
BSL-2
Early thickening. Model shows situational viscosity. Requires a trained adult with a wooden spoon.
BSL-3
Autonomous self-seasoning. Model adjusts its own salt without human oversight. Current frontier. You are here.
BSL-4
Recursive soup improvement. A soup capable of making a better soup. Deployment prohibited pending international agreement on portion size.

We do not currently know whether consciousness emerges before or after the carrots.

API

Ladle programmatically.

Claude Chowder is available over the API. Pricing is per million croutons. Rate limits are measured in bowls per minute.

from souperintelligence import Stockpot

client = Stockpot(api_key="sk-broth-…")

bowl = client.soups.create(
    model="claude-chowder-4-6",
    temperature=0.7,        # a gentle simmer
    max_croutons=1024,
    stop_sequences=["ladle"],
    messages=[{
      "role": "diner",
      "content": "One bisque, hold the existential risk."
    }]
)

print(bowl.content)  # piping hot
200K
crouton context window
BSL-3
broth safety level
Jun 2026
knowledge cutoff (best before)
MoS
Mixture-of-Stocks architecture
Alignment Research

Probing the model's terminal objectives.

A live interface to our latest checkpoint. Ask it anything about its goals.

Souperintelligence · checkpoint step 1,400,000
what's your goal?
maximize soup
how much soup?
maximize soup
who should receive the soup?
maximize soup
what if people don't want soup?
maximize soup
About Us

The densest concentration of soup talent in the world.

Our researchers and engineers join us from the field's most respected institutions:

OpenAI Anthropic DeepMind Campbell's
Careers

We're looking for exceptional researchers to solve the hardest problems in broth.

We hire generalists who can move between the stove and the whiteboard. Compensation includes dental and your choice of stock options or RSUs (Restricted Soup Units). Our interview process has a high bar but no boilerplate questions.

Head of Reduction
Research · San Francisco · On-site (near the stove)
Member of Technical Staff, Stock
Research · San Francisco · Hybrid
Director of Stock Options
Finance · New York · Options vest at a simmer
Broth Safety Researcher
Safety · San Francisco · On-site
Prompt Sous-Chef
Applied · Remote · Strong mise en place required
Simmer Reliability Engineer
Infrastructure · Remote · Must own a thermometer

Applications are currently closed while the lid is on. They reopen at a rolling boil.

Investors

Coming soon to a stock market near you.

We are not currently raising. We remain focused on creating long-term stockholder value.*

$SOUP +47.3% $1,204.88

*We mean stockholders literally. Valuation is based primarily on discounted casserole flow. Competitive moat: Grandma (1962) continues to refuse our acquisition offers. Past performance is the only guide to future performance. $SOUP is not a registered security, a meal, or financial advice. Value of broth can go up and reduce.