Skip to content
Oumi AI

Atlas of AI

Specialized models for every business job, ranked and ready to build.

Find AI worth owning

580 Specialist Models, one per business decision.

54 of 580 models

#73 overall · #2 in Cross-industry

Documentation Generation

What technical documentation (API reference, code comments, README, runbook) should be generated or updated for this codebase change, and is it accurate enough to publish?

Owned by VP Engineering, Head of Developer Experience

Cross-industry
Build it: Documentation GenerationRead the guide: Documentation Generation

#74 overall · #3 in Cross-industry

Expense Audit

Should this expense report or payment be approved, held for review, or escalated as suspected fraud or policy violation?

Owned by Controller, VP Finance Operations

Cross-industry
Build it: Expense AuditRead the guide: Expense Audit

#107 overall · #8 in Cross-industry

URL Risk

Is this URL malicious or phishing, and what action should be taken (block, warn, allow)?

Owned by CISO, Head of Security Operations

Cross-industry
Build it: URL RiskRead the guide: URL Risk

#160 overall · #9 in Cross-industry

Benchmark Coverage

Which real-world capability gaps in an AI system are NOT covered by its current benchmarks/eval sets, and what new eval cases should be added to close those gaps?

Owned by Head of AI, ML Platform Lead

Cross-industry
Build it: Benchmark CoverageRead the guide: Benchmark Coverage

#172 overall · #10 in Cross-industry

Competitive Messaging

Given a specific competitor and deal context, what positioning, proof points, and objection-handling language should a rep or marketer use right now?

Owned by VP Product Marketing, Head of Competitive Intelligence

Cross-industry
Build it: Competitive MessagingRead the guide: Competitive Messaging

#175 overall

Context Ranking

Given a user query/task and a large pool of retrieved candidate documents or context chunks, which should be ranked highest and passed into the model's context window?

Owned by Head of AI, VP Engineering

Cross-industry
Build it: Context RankingRead the guide: Context Ranking

#194 overall

Evidence Selection

Given a set of retrieved candidate evidence spans or documents, which specific evidence should be selected and cited to support a generated answer?

Owned by Head of AI, VP Product (Enterprise Search/Assistant)

Cross-industry
Build it: Evidence SelectionRead the guide: Evidence Selection

#203 overall

Fine-Tune Candidate

Which production tasks or prompt clusters currently running on a frontier model are good candidates to move to a fine-tuned specialist model, based on volume, cost, latency, and quality-gap signals?

Owned by Head of AI, ML Platform Lead

Cross-industry
Build it: Fine-Tune CandidateRead the guide: Fine-Tune Candidate

#258 overall

SOC Copilot

For this alert, what is the likely severity and root cause, and what investigation or response actions should the analyst take next?

Owned by CISO, SOC Manager

Cross-industry
Build it: SOC CopilotRead the guide: SOC Copilot

Choose a model you can own.

Model leaderboards are a starting point. The decision is your task, your output contract, your hardware, and your right to deploy. This directory turns that into a short list you can test.

  • Qwen

    Alibaba Cloud · China
    Qwen is Alibaba Cloud’s open-weight model family. It spans small models that are practical to fine-tune through larger mixture-of-experts releases.
    Active in Oumi

    Profile: Qwen

  • DeepSeek

    DeepSeek · China
    DeepSeek publishes open-weight language models, including large mixture-of-experts systems. The family is often shortlisted where reasoning and coding matter.
    Listed by Oumi

    Profile: DeepSeek

  • Llama

    Meta · United States
    Llama is Meta’s open-weight model family. Its wide ecosystem makes it a familiar baseline for teams evaluating or fine-tuning their first custom model.
    Listed by Oumi

    Profile: Llama

  • Mistral

    Mistral AI · France
    Mistral is a European model family spanning open-weight and commercial releases. The right starting point depends on the exact release and deployment constraints.
    Listed by Oumi

    Profile: Mistral

  • Gemma

    Google · United States
    Gemma is Google’s open-weight model family. It includes compact checkpoints that can be useful when hardware limits and deployment footprint drive the decision.
    Listed by Oumi

    Profile: Gemma

Oumi benchmark data is coming. We’ll add reproducible task benchmarks once the evaluation recipe, model version, dataset, hardware, and scoring criteria are published.

Find the AI your company should own.

Enter your work email. Get a shortlist of Specialist Models matched to your company, in about 30 seconds.

Find AI worth owning

Enter your work email · We work out your industry · Get your shortlist, then build one

Questions about the Atlas

  • What is the Atlas of AI?

    The Atlas of AI is a directory of specialist-model opportunities organized around business decisions and tasks. It helps you explore where a focused model may create value, then move toward evaluation and ownership.

  • How do I find a Specialist Model for my business task?

    Start with the business decision or repeated workflow you want to improve, then use Atlas filters and rankings to narrow the relevant category. Treat the result as a discovery hypothesis and validate it against your data, evaluation criteria, and deployment needs.

  • Which open-weight model should I start from?

    Start with the open-weight model that fits your task, quality target, context needs, license, hardware, and deployment constraints. There's no single best model for every workflow, so use Atlas evidence and your own evaluation set.

  • What evidence supports each Specialist Model listing?

    A listing is a structured starting point grounded in the task, model, benchmark, or other evidence shown with it. Review the evidence and run your own evaluation before treating a listing as a production recommendation.

  • Can I build a Specialist Model directly from the Atlas?

    The Atlas helps identify a promising task and starting model. From there, use Oumi's build and evaluation workflow to define behavior, assemble data, test the model, and decide whether to train and deploy it.

Related items

2026 State of Owned AI Report

Report · PDF

Read the report
Murali Minnah, webinar speaker

The Business Case for Owning Your AI

On-demand webinar · 42 min

Watch now
Monthly savings with Oumi: $205,560, 96.7% lower than renting an API at $212,500, Oumi at $6,944.40

What is frontier AI actually costing you?

Rent vs Own Calculator

Run your numbers
SeeWatch on-demand demosLearnExplore resourcesMeetTalk to an expert

Pick a Specialist Model. Oumi plans the build.

Start free in the platform, or talk to us about your task.

Get started freeBook a demo