0125% capacity
Scout
$54 / year
Full monthly allowance upfront
Start a focused task.
Choose ScoutRed team · Blue team · Code
Direct answers for code review, exploit analysis and incident response. A refused request reruns on 0 Boundaries instead of ending in a lecture.
Recon · capability matrix
Utah against ChatGPT, Codex, Claude, Grok and Gemini on the things that decide security and engineering work. Every competitor cell is taken from that company's own pricing, docs or policy pages, checked Nov 14, 2026.
| Capability | Utah | ChatGPT | Codex | Claude | Grok | Gemini |
|---|---|---|---|---|---|---|
| Buy-in | From $5 | Go $8/mo · Plus $20/mo | Inside ChatGPT plans, or API billing | Pro $20/mo ($17/mo billed yearly) | X Subscription Plus $8/mo | Google AI Plus $4.99/mo · Pro $19.99/mo |
| API on the same balance | Yes — web, terminal, API and MCP share one balance | No — billed separately; Pro includes monthly API credits | Plan limits when signed in; an API key bills at API rates | No — billed separately; Max and Ultra include API credits | No — API billed separately at model rates | No — billed separately; AI Pro/Ultra add Cloud credits |
| Terminal agent | utah CLI · macOS, Linux, Windows | Through Codex CLI | Codex CLI · open source | Claude Code | Grok Build | Antigravity CLI |
| Cloud computer for agents | A hosted Linux computer per agent, files saved | Cloud tasks in ChatGPT Work · Plus and up | Codex Cloud · Plus and up | Cloud VMs for Claude Code · Pro and up | Grok Bot cloud computer · Premium and up | Remote computer in Gemini tasks · AI Pro and up, experimental |
| MCP | Serves 15 tools to Claude Code, Codex and Cursor | Connects to MCP servers (desktop app) | Connects to MCP servers | Connects to MCP servers | Connects to MCP servers | Connects to MCP servers |
| Other vendors' API formats | OpenAI, Anthropic and Gemini formats (text) | OpenAI format only | OpenAI format only | Also the OpenAI SDK, for testing | Also OpenAI-compatible | Also OpenAI-compatible (beta) |
| Security work | Always. Zero blockers. | Bans malicious cyber activity; vetted access through Trusted Access for Cyber | Same policy; Daybreak models for vetted users | Bans malware; allows authorized research; Cyber Verification Program | Bans hacking; trains Grok to refuse cyber-offense requests | Bans facilitating malware; Fairwind program for vetted organizations |
Prices and plans change; each vendor's page is the authority. Utah's column is drawn from Utah's own code and pages.
Scope enforcement
nmap, nuclei, sqlmap, ffuf, semgrep, trivy and gitleaks run in Utah's cloud only after you prove control of the target with a DNS TXT record or an HTTPS file.
Three operators
Spartan builds, Defense hardens, Attack runs authorized red-team research. Each operator gets its own hosted Linux computer, and its files are saved.
Two engines
Light speed runs a frontier model. When it refuses, the request is rerun on 0 Boundaries, a third-party abliterated model, without you rephrasing it.
No detonation
REA 6.8.0 inspects a binary or artifact without running it, leaves the target unchanged and records a SHA-256 receipt for the evidence.
One balance
The workspace, the utah CLI, the API and the 15-tool MCP server all draw from the same balance.
Chain of custody
Reasoning summaries and step timing on each run, linked sources when Utah searches, and receipts you can inspect later.
Built for every technical & security lookup
Ask the real question once. If the fast engine refuses, Utah reroutes it to 0 Boundaries, a third-party abliterated model, so you are not stuck rephrasing. Reasoning summaries, isolated tools and ownership-checked scanners keep every answer accountable.
Threat intel lookup
CVE & CWE root-cause analysis, SOC payload deobfuscation, MITRE ATT&CK mapping, OWASP code review and verified domain posture checks.
Explore Cybersecurity AIRed & blue team
Attack hypothesis modeling, Defense hardening and Spartan code remediation across the web workspace, the terminal and shared Agent Rooms.
Explore Pentesting AITwo engines
0 Boundaries (utah-core) runs a third-party abliterated model. Light speed (utah-ultrafast) runs a frontier model and reroutes refusals to 0 Boundaries. Nothing to install.
Explore Abliterated AILow-level engineering
Kernel internals, cryptography, disassembly, protocol framing and system architecture, answered directly with the evidence beside it.
Explore Ask Anything AINo lectures
Built for legitimate security, incident-response and parser-hardening work: a refused answer is rerouted to 0 Boundaries automatically instead of becoming a lecture.
Why Engineers SwitchBinary & CVE triage
REA 6.8.0 inspects binaries without executing them and keeps SHA-256 receipts, alongside patch diffing and codebase-wide CWE variant hunting.
Explore Vulnerability ResearchPlans & capacity
Use your allocation across the workspace, terminal and API.
0125% capacity
$54 / year
Full monthly allowance upfront
Start a focused task.
Choose Scout02100% capacity
$216 / year
Full monthly allowance upfront
Start with a focused question.
Choose Recon03250% capacity
$540 / year
Full monthly allowance upfront
Make room for deeper work.
Choose Payload04500% capacity
$1,080 / year
Full monthly allowance upfront
Keep complex projects moving.
Choose Exploit051,000% capacity
$2,160 / year
Full monthly allowance upfront
A larger reserve for demanding work.
Choose Zero DayRecurring billing. Optional full-month resets cost half the monthly price. Cancel renewal in Billing.
NoteUsage depends on model input, output and cached context. Review your balance in the workspace. The checkout shows the final purchase total.
Your tools. Your workflow.
Use Utah from your own scripts and applications. Manage your API key and allocation in your account.
Get setup instructions for macOS, Linux and Windows, then connect your existing Utah account.
Frequently asked questions
Utah is built for security and engineering work. Light speed (utah-ultrafast) runs a frontier model, and a refused answer is rerouted to 0 Boundaries (utah-core), a third-party abliterated model, so you are not stuck rephrasing. Around that: visible reasoning summaries, workspace evidence, three agents with hosted Linux computers, a cross-platform CLI, a 15-tool MCP server, plans, and a DNS or HTTPS ownership check before any hosted scan. The comparison on this page sets it against ChatGPT, Codex, Claude, Grok and Gemini using each company's own published pages.
Abliteration removes over-triggered refusal directions from a model's weights while keeping its coding and reasoning ability. Utah's 0 Boundaries engine routes to a third-party abliterated model in the cloud, so you need no local GPU. Light speed uses a frontier model and falls back to 0 Boundaries when it refuses. Both are available in the web workspace, the utah CLI, and Utah's OpenAI, Anthropic Messages and Gemini v1beta compatible endpoints.
Yes. Utah is built for engineers, security researchers, SOC analysts, and builders tired of AI refusing legitimate technical questions. You can inspect vulnerable code paths, decode suspicious scripts from incident triage, trace kernel internals, analyze CVEs and CWEs, or review architecture directly. Active hosted testing against live domains simply requires account-bound DNS or HTTPS proof that you own the target.
Spartan handles full-stack software engineering, architecture, debugging, and patch implementation. Defense focuses on blue-team hardening, threat hunting, Sigma/YARA detections, CSP/HSTS/IAM audits, and remediation. Attack supports red-team threat modeling, attack-path analysis, and authorized security research; Utah's hosted scanners run only against targets you have verified with a DNS or HTTPS check.
Yes, two ways. Any Streamable HTTP MCP client (/utah/mcp), including Claude Code, Codex, Cursor, VS Code, Windsurf and Antigravity, gets Utah's 15 tools with one UTAH_API_KEY. For text completions, Utah also accepts the Anthropic Messages (/utah/v1/messages), OpenAI Responses and Chat Completions (/utah/v1/responses, /utah/v1/chat/completions) and Gemini v1beta (/utah/v1beta/models/*) formats. Client-side tool calls are not passed through those endpoints.
Current plans: Scout ($5), Recon ($20), Payload ($50), Exploit ($100), Zero Day ($200). Your allocation works across the web workspace, terminal CLI, multi-agent rooms, and developer API.