Hero Background Image

New

Crucible 3.0 is here

The reasoning model
built for what matters.

Crucible is an AI model that works through problems carefully before producing an answer. Connect via API, build in minutes, and ship intelligence you can actually trust.

Analyze my contract

Summarize Art of War

Review my code

Analyze my contract

Summarize Art of War

Review my code

Trusted by teams building serious products

One model with two ways to build.

Both connect to the same underlying model.

Card Background Image

import crucible

client = crucible.Client(

api_key="sk-cru-••••••••••••"

)

response = client.reason(

model="crucible-pro",

messages=[{

"role": "user",

"content": prompt

}],

output_format="json"

)

import crucible

client = crucible.Client(

api_key="sk-cru-••••••••••••"

)

response = client.reason(

model="crucible-pro",

messages=[{

"role": "user",

"content": prompt

}],

output_format="json"

)

200 OK

200 OK

743 ms

743 ms

Crucible API

Connect directly to the model via our REST API. Full control over every parameter, output format, and model tier. Python and TypeScript SDKs included.

Access API

Card Background Image

  • Hey do you know how many planets are there in our galaxy? make a list based on their sizes, also make it as a table.

    Ask questions

    Sure, taking a look...

  • Build interface

    Create a landing page for SaaS


    "dashboard">


    Risk Overview


    Monitor key risks across all quarters


    View Report



  • Compare models

    Standard

    The report states that patient outcomes improved and...

    Select Output 1

    Card Background Image

    Pro

    Patient outcomes show improvement and the treatment...

    Select Output 2

    Which output is better?

  • File uploads + scan

    Scanning files...

    Illustration Item

    Certific...

    Illustration Item

    Give me a quick summary

Crucible Playground

Test prompts, compare model outputs, and prototype before you ship. No code required to get started.

Open Playground

Card Background Image

Q4_Report-2077

Study_2026.pdf

idea-dump.md

saas-website.md

Q4_Report-2077

Study_2026.pdf

idea-dump.md

saas-website.md

Q4_Report-2077.xlxs

Study_2026.docx

Primary findings:

34% reduction in processing time

10X better output compared to Pro 1

Idea_PRD.pdf

Fintech App Idea PRD

saas-website.html

Q4_Report-2077.xlxs

Study_2026.docx

Primary findings:

34% reduction in processing time

10X better output compared to Pro 1

Idea_PRD.pdf

Fintech App Idea PRD

saas-website.html

Feed it anything. Get back what matters

Feed it anything. Get back what matters

Most models read a document and summarize it. Crucible reads a document and reasons through it. It distinguishes between what a clause says and what it means. Between what a report states and what the data actually supports.

Most models read a document and summarize it. Crucible reads a document and reasons through it. It distinguishes between what a clause says and what it means. Between what a report states and what the data actually supports.

Contracts

Research Papers

Financial Filings

Internal Memos

Product Specifications

Ask Questions

Extract Information

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%

Factual accuracy on structured queries

Factual accuracy on structured queries

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
X
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
X

More citations per output vs baseline

More citations per output vs baseline

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9

Page maximum document length (Pro)

Page maximum document length (Pro)

Get answers in the format that’s expected.

Get answers in the format that’s expected.

Crucible returns structured data by default. Pass a JSON schema and the model validates its output against it before responding. No post-processing. No regex. No brittle parsing. Tables, lists, nested objects, or plain text. Every output is formatted for downstream use, whether you are feeding it into a database, a dashboard, or another model in your pipeline.

Crucible returns structured data by default. Pass a JSON schema and the model validates its output against it before responding. No post-processing. No regex. No brittle parsing. Tables, lists, nested objects, or plain text. Every output is formatted for downstream use, whether you are feeding it into a database, a dashboard, or another model in your pipeline.

Open Playground

Card Background Image

Python

Typescript

cURL

import crucible

client = crucible.Client(

api_key="sk-cru-••••••••••••",

)

response = client.reason(

model="crucible-pro",

messages=[{

"role": "user",

"content": "Extract key terms from this agreement.",

}],

output_format="json",

)

Python

Typescript

cURL

import crucible

client = crucible.Client(

api_key="sk-cru-••••••••••••",

)

response = client.reason(

model="crucible-pro",

messages=[{

"role": "user",

"content": "Extract key terms from this agreement.",

}],

output_format="json",

)

Card Background Image

auth.py

auth.py

def authenticate(token):

return valid_token(token)

def authenticate(token):

return valid_token(token)

payments.py

payments.py

if is_payment_valid(data):

process_payment(data)

if is_payment_valid(data):

process_payment(data)

Reasoning

Reasoning

Authentication failure propagates to payment validation¹

Missing token check in payments.py²

Leads to inconsistent API behavior

Authentication failure propagates to payment validation¹

Missing token check in payments.py²

Leads to inconsistent API behavior

Get actual reasoning about your code.

Get actual reasoning about your code.

Crucible reads your existing codebase, not just the function you paste in. It understands context, traces logic across files, and identifies issues that only appear when you look at the whole picture. Code review, debugging, refactoring suggestions, and plain-language explanations of complex functions. Each suggestion includes a clear reason, not just a replacement.

Crucible reads your existing codebase, not just the function you paste in. It understands context, traces logic across files, and identifies issues that only appear when you look at the whole picture. Code review, debugging, refactoring suggestions, and plain-language explanations of complex functions. Each suggestion includes a clear reason, not just a replacement.

Open Playground

Everything a production integration needs.

Uptime you can build on

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Uptime you can build on

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Uptime you can build on

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Latency that works in production

Average first token under 800ms on Pro. Sub-400ms on Standard for lighter, lower-complexity tasks.

Latency that works in production

Average first token under 800ms on Pro. Sub-400ms on Standard for lighter, lower-complexity tasks.

Latency that works in production

Average first token under 800ms on Pro. Sub-400ms on Standard for lighter, lower-complexity tasks.

No training on your data

Your prompts and responses are never used to train Crucible models. Encrypted in transit and at rest. SOC 2 Type II certified.

No training on your data

Your prompts and responses are never used to train Crucible models. Encrypted in transit and at rest. SOC 2 Type II certified.

No training on your data

Your prompts and responses are never used to train Crucible models. Encrypted in transit and at rest. SOC 2 Type II certified.

Limits that match real workloads

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Limits that match real workloads

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Limits that match real workloads

99.5% uptime SLA on Pro. 99.99% on Enterprise. Status page updated in real time at status.crucible.ai.

Confidence is not the same as being correct.

Confidence is not the same as being correct.

We publish our benchmark results because accuracy that cannot be measured cannot be trusted. Crucible Pro 2.1 achieves 94.1% factual accuracy on our structured document analysis benchmark, a dataset of 2,400 legal and financial tasks with verified ground truth. Every response includes citations. When Crucible draws from a source in your document, it tells you exactly where. You can verify every claim before it leaves your pipeline.

We publish our benchmark results because accuracy that cannot be measured cannot be trusted. Crucible Pro 2.1 achieves 94.1% factual accuracy on our structured document analysis benchmark, a dataset of 2,400 legal and financial tasks with verified ground truth. Every response includes citations. When Crucible draws from a source in your document, it tells you exactly where. You can verify every claim before it leaves your pipeline.

Read the research

Card Background Image

Verified Output

Verified Output

The report shows consistent revenue growth across all quarters¹ with operating margins improving steadily²

The report shows consistent revenue growth across all quarters¹ with operating margins improving steadily²

Page 4, Section 2

Page 4, Section 2

Page 11, Financial Summary

Page 11, Financial Summary

Every claim is grounded in source evidence

Every claim is grounded in source evidence

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%

Factual accuracy on document analysis

Factual accuracy on document analysis

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
%

Reduction in hallucination rate, Pro 2.1

Reduction in hallucination rate, Pro 2.1

0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9

Tasks in our verification benchmark

Tasks in our verification benchmark

Latest Updates

Read our latest update

How we reduced hallucination rates by 12% in Crucible Pro 2.1

Learn how Crucible Pro 2.1 improved accuracy to 94.1% and reduced hallucinations through smarter reasoning.

Engineering

Apr 3, 2026

How we built Crucible's streaming pipeline

A technical walkthrough of the server-sent events architecture that powers real-time output in Crucible.

Engineering

Feb 18, 2026

Designing for latency: the engineering behind Fast Mode

Inside the architectural decisions that make Fast Mode fast without sacrificing reliability or performance.

Engineering

Dec 10, 2025

Read all

Simple, transparent pricing.

Start free. Scale with usage-based pricing when you're ready. No contracts for Pro. Custom terms for Enterprise.

Monthly

Yearly

Free

For developers getting started

$0/month

$0/month

100K tokens per month

Crucible Standard Model

32K token context window

60 requests per minute

REST API access

Text and Markdown output

Start with Free

Pro

For teams building in production.

$49/month

$49/month

10M tokens per month

$0.80 per additional 1M tokens

Standard + Pro model

128K token context window

500 requests per minute

JSON output mode + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.5% uptime SLA

Email support, 48h response

30-day audit logs

Start with Pro

Scale

For high-volume production teams.

$199/month

$199/month

50M tokens per month

$0.60 per additional 1M tokens

Standard + Pro + Giga model

200K context window

2,000 requests per minute

JSON output + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.9% uptime SLA

Priority support, 8h response

90-day audit logs

Start with Scale

Enterprise

For teams with specific requirments

Custom

Custom token limits

SSO and access controls

Dedicated support

Custom SLA

Contact Sales

Monthly

Yearly

Free

For developers getting started

$0/month

$0/month

100K tokens per month

Crucible Standard Model

32K token context window

60 requests per minute

REST API access

Text and Markdown output

Start with Free

Pro

For teams building in production.

$49/month

$49/month

10M tokens per month

$0.80 per additional 1M tokens

Standard + Pro model

128K token context window

500 requests per minute

JSON output mode + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.5% uptime SLA

Email support, 48h response

30-day audit logs

Start with Pro

Scale

For high-volume production teams.

$199/month

$199/month

50M tokens per month

$0.60 per additional 1M tokens

Standard + Pro + Giga model

200K context window

2,000 requests per minute

JSON output + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.9% uptime SLA

Priority support, 8h response

90-day audit logs

Start with Scale

Enterprise

For teams with specific requirements.

Custom

Custom token limits

SSO and access controls

Dedicated support

Custom SLA

Contact Sales

Monthly

Yearly

Free

For developers getting started

$0/month

$0/month

100K tokens per month

Crucible Standard Model

32K token context window

60 requests per minute

REST API access

Text and Markdown output

Start with Free

Pro

For teams building in production.

$49/month

$49/month

10M tokens per month

$0.80 per additional 1M tokens

Standard + Pro model

128K token context window

500 requests per minute

JSON output mode + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.5% uptime SLA

Email support, 48h response

30-day audit logs

Start with Pro

Scale

For high-volume production teams.

$199/month

$199/month

50M tokens per month

$0.60 per additional 1M tokens

Standard + Pro + Giga model

200K context window

2,000 requests per minute

JSON output + custom schemas

Citation mode

Batch API access

Prompt caching

Extended reasoning mode

99.9% uptime SLA

Priority support, 8h response

90-day audit logs

Start with Scale

Enterprise

For teams with specific requirements.

Custom

Custom token limits

SSO and access controls

Dedicated support

Custom SLA

Contact Sales

Common Questions

Some queries we usually get

How does token-based pricing work?

Tokens are the units Crucible uses to measure text. Your prompt and Crucible's response both count toward your monthly total. As a rough guide, 1,000 tokens is about 750 words. On Pro, additional tokens beyond 10M are billed at $0.80 per million.

Can I switch plans at any time?

Yes. You can upgrade or downgrade at any time from your account settings. Upgrades take effect immediately and are prorated. Downgrades apply at the start of your next billing cycle.

What happens if I exceed my monthly token limit?

On the Free plan, generation pauses until your limit resets at the start of the next month. On Pro, overages are billed automatically at $0.80 per million tokens, so your work is never interrupted.

Do you offer annual billing?

Yes. Annual plans are billed upfront at a 20% discount compared to monthly. You can switch to annual at any time from your account settings. No extra steps required.

Is there a free trial for Pro?

Every new account starts on the Free plan with no credit card required. You can explore the core reasoning features before committing. When you're ready to upgrade, it takes about 30 seconds.

Does Crucible train on my data?

No. Your inputs and outputs are never used to train or fine-tune any model. Data submitted through the API is not retained beyond your request unless you explicitly enable memory features.

What models are available on each plan?

The Free plan includes access to Crucible Standard. Pro unlocks Crucible Extended for deeper reasoning tasks, plus priority access to new model releases as they ship.

What is citation mode?

Citation mode instructs Crucible to reference the specific sources or passages its reasoning draws from. It is designed for research workflows, compliance use cases, and any context where verifiable, traceable outputs matter.

Can I use Crucible without writing code?

Yes. The Crucible interface is fully browser-based and requires no setup. The API is available for teams who want to integrate reasoning into their own products, but it is entirely optional.

Get Template

Create a free website with Framer, the website builder loved by startups, designers and agencies.