TypeSafe AI overview: products, pricing, and alternatives
Author
Aanchal Tiwari
Date Published
Reading Time
5 min read
-600x315-R0tAjzfNUNSE0HsoA0U9M9aopFEoT7.webp%3F2026-10-30T19%253A03%253A17.860Z&w=3840&q=82&dpl=dpl_5qejmRFWDJdrBWZYBSRGaSucinzS)
TypeSafe AI, the company behind Jev, is an AI lab that builds decision models: Jev reads text or JSON and returns typed decisions instead of prose. Founded in San Francisco in 2024 by InstructGPT co-author Diogo Almeida, Erik Gafni, and Sasha Sheng, it raised about $910M and serves developers automating high-volume decisions.
TypeSafe AI at a glance
Field | Value |
Founded | 2024 |
Headquarters | San Francisco, United States |
Founder / CEO | Diogo Almeida (CEO), Erik Gafni, Sasha Sheng |
Ownership | Independent, VC-backed |
Funding raised | About $910M (latest: $870M Series A, Oct 2026) |
Category | |
Flagship product | Jev |
Starting price | $0.042 per 1M input tokens (output free) |
Free plan | No |
Website | typesafe.ai |
What is TypeSafe AI?
TypeSafe AI is a “System One” model maker, named after Kahneman’s fast-thinking System 1, whose output is a decision for software to use rather than text for people to read. Its only model, Jev, takes text or JSON plus questions defined in code and returns a Choice, a rubric Score, a Noul (yes/no probability), or an experimental Box, with no rationale. Teams use it for classification, routing, scoring, guardrails, and agent control inside their own software.
TypeSafe AI products
- Jev: The flagship System One decision model; current versions are jev-1.14.0 and jev-1.13.0.
- System One API: One endpoint, POST /v1/systemone, with Choice, Score, Noul, and Box question types.
- Official SDKs: Python (typesafe-sdk) and JavaScript (@typesafe-ai/sdk), both MIT-licensed.
- Enterprise offering: Zero data retention, SOC 2 report access, and higher rate limits through sales.
TypeSafe AI pricing
Jev costs $0.042 per million input tokens, and new direct signups get no free tier or credit. Billing is pay-as-you-go on input tokens only: output is free, long context costs the same per token, and there are no plans or subscriptions. The limits readers hit first are per-organization rate limits and the 64k-token default context. TypeSafe has no dedicated pricing page, so verify current rates on docs.typesafe.ai/models.
Data, privacy, and safety
No, TypeSafe AI does not train or fine-tune its models on your prompts, inputs, or outputs, per its privacy policy and enterprise FAQ. Zero data retention is offered to enterprise customers only, and the service is hosted in the United States.
Certifications: SOC 2 Type II; DPA with EU SCCs and UK Addendum; HIPAA/BAA, ISO 27001, and VPC/on-prem: Not publicly specified.
No confirmed breach was found. HiddenLayer’s limited October tests did show planted content could steer Jev’s choices in a simulated store.
Who’s behind TypeSafe AI and how stable it is
TypeSafe AI is heavily funded and shipping actively, most recently Jev 1.14.0 on October 14, 2026. The independent company raised a $40M seed led by DCVC, then an $870M Series A on October 9, 2026, led by Andreessen Horowitz with Sequoia, and a16z’s Martin Casado joined the board. CEO Diogo Almeida leads with CTO Erik Gafni and COO Sasha Sheng. Almeida’s claim to Bloomberg that TypeSafe is profitable is unverified, and signups paused for about a week in September under demand.
TypeSafe AI strengths and limitations
Strengths
- Charges $0.042 per million input tokens with free output, and quotes 70 to 500 ms responses.
- Publishes a jaggedness page listing where Jev fails, which a Pinggy review called unusually candid.
- Promises no training on inputs, holds SOC 2 Type II, and commits to 72-hour breach notice in its DPA.
- Available through OpenRouter, Vercel AI Gateway, Cloudflare Workers AI, MotherDuck SQL, and Braintrust.
- Jev 1.13 has processed 3.39T tokens on OpenRouter, and Workday’s Sana assistant uses it.
Limitations
- Cannot write text, code, or explanations, so developers wanting an AI coding assistant or chatbot should look elsewhere.
- Trails Liquid AI d1-ultra on an Artificial Analysis Decision Index mirror, 57.1 vs 65.2 (Oct 19 snapshot).
- TypeSafe’s own docs admit weak numeric precision, dates, counting, and overly literal question reading.
- Teams needing EU processing or self-hosting should look elsewhere: weights are closed and the API is US-only.
- Image input is experimental, and Fortune 500 adoption claims (25% to one third) name no companies.
TypeSafe AI alternatives
OpenAI Decisions API, Microsoft-Decision-1, Liquid AI d1-ultra, and AWS Strands Decider 2B are TypeSafe AI’s closest alternatives. TypeSafe still has the deepest dedicated toolkit, but Microsoft and Liquid AI now charge the same $0.042, and d1-ultra scores higher.
Company | What they do | Best for | Entry price | Key difference vs TypeSafe AI |
TypeSafe AI (Jev) | Decision model returning choices, scores, yes/no | High-volume routing and guardrails | $0.042 per 1M input tokens, pay-as-you-go | Baseline |
OpenAI Decisions API | Typed answers from text, images, JSON (gpt-6-luna) | OpenAI users needing images or EU processing | $0.10 per 1M uncached input tokens, pay-as-you-go; cache writes charged | About 2.4x higher input price; general LLM, not decision-only |
Microsoft-Decision-1 | Decision model on Qwen3.5-9B in Microsoft Foundry | Azure and Foundry customers | $0.042 per 1M input tokens (US and EU Datazone) | Public preview, Azure only, 32,768-token text-only context |
Liquid AI d1-ultra | Liquid’s largest decision model, text and images | Image decisions, top benchmark score | $0.042 per MTok input, $0.00 output; $2 free credit | Each question billed as its own prompt, raising image costs |
AWS Strands Decider 2B | Free open 2B decision model you host | Local or on-prem, zero per-call cost | Free (Apache 2.0, self-hosted) | 4,096-token limit, self-hosted, experimental |
Frequently Asked Questions
Does TypeSafe AI train on my data?▾
Can Jev write text or code?▾
Is Jev free to use?▾
Did TypeSafe AI have a data leak?▾
What is TypeSafe AI’s valuation?▾
Is TypeSafe AI hiring?▾
Written by
Aanchal Tiwari
Aanchal Tiwari is an AI tools reviewer and content writer with 3 years of experience breaking down software platforms, pricing models, and product capabilities. A B.Com graduate from SPM College (Delhi University) with a diploma in office management, she brings a structured, detail-first approach to every company overview she publishes at FirmBees. Her process starts with a real walkthrough of the product and its fine print, so readers know exactly what they are paying for before they commit.
Comments
Leave a comment
Related posts
-600x315-dG44fJG4U1NhwpQvl2Iq5mNyGTN5lc.webp%3F2026-10-05T19%253A34%253A16.689Z&w=3840&q=82&dpl=dpl_5qejmRFWDJdrBWZYBSRGaSucinzS)
StackBlitz overview: the maker of Bolt.new and its default-on training policy
StackBlitz, Inc. is the San Francisco company behind Bolt.new, an AI app builder that runs Node.js code inside the browser tab. Since October 7, 2026, Bolt content from individual Free and Pro accounts can train its models and be licensed as datasets unless users opt out.
-jeuG1j73yiYO6ivyDir4deqIJ376oy.webp%3F2026-08-28T13%253A16%253A21.232Z&w=3840&q=82&dpl=dpl_5qejmRFWDJdrBWZYBSRGaSucinzS)
Wix overview: products, pricing, and alternatives
Wix.com Ltd. is a publicly traded (Nasdaq: WIX) website builder used by small businesses and creators, offering AI site creation through Wix Harmony, a free plan, and paid tiers from $19 a month, alongside its 2025 Base44 acquisition.

Lovable overview: products, pricing, and alternatives
Lovable is a Swedish AI platform that turns plain-language prompts into real React and Supabase web apps. Valued at $13.3 billion after a $400M Series C in August 2026, it targets non-technical founders and product teams.