Gains:
- Matching strong fields and typical user of Claude, GPT and Gemini families
- Ability to read the level (tier) logic and model naming of each provider
- Get in the habit of basing comparisons on your own testing data, not marketing claims
We mapped the market and learned the closed/open weight distinction. Now we will get to know the three most visible players in the market, namely the large US-based providers: Anthropic (Claude), OpenAI (GPT) and Google (Gemini). These three feature in the majority of enterprise AI decisions today. Our goal is not to declare a "winner"; The aim is to objectively understand where each one shines, its naming logic and its typical user. When the unit is finished, you will be able to talk about these three families without confusing them and you will internalize that you should base a comparison on your own test, not on advertising.
Common Logic: Tier System
These three providers all use a similar logic: they offer models within the same family, tiered according to the speed/cost/power balance. There are usually three tiers:
- Light tier: Fastest and cheapest. For simple, high-volume tasks (classification, briefing, labeling).
- Balanced stage: Mid-range power and cost. The "default" choice for most daily tasks.
- Strong tier: Most capable, most expensive, slowest. For complex reasoning, long analysis, difficult code.
Once you understand this tier logic, you won't feel lost no matter which provider you look at. "Which level corresponds to this model in this family?" It is enough to ask.
Tip: When you hear about a new model, first ask "which level of which family?" ask. Don't let the multitude of names scare you; Each provider actually offers the same three tiers under different names.
Anthropic — Claude Family
Anthropic's Claude family offers three tiers called Opus (strong), Sonnet (balanced), and Haiku (light). Version numbers are also included; For example, the model used by this platform is claude-opus-4-8, that is, the 4.8 version of the Opus stage.
It excels in: long context processing (the ability to read hundred-page documents in their entirety), writing and reading code, and secure, predictable behavior in enterprise use. Claude is known for staying close to instructions and acting poised on sensitive matters. Therefore, it is frequently preferred in regulated sectors such as law, finance and health.
Typical user: Legal teams analyzing long documents, software teams reviewing code, corporate customer services requiring secure output.
Current Anthropic models and approximate prices (per million tokens, input/output):
model
Stage
Context
Input $/1M
Output $/1M
Claude Opus 4.8
strong
1M tokens
~5
~25
Claude Sonnet 5
balanced
1M tokens
~3
~15
Claude Haiku 4.5
lightweight
200K tokens
~1
~5
Attention: These prices are approximate for the period in which this module was prepared and change frequently. Be sure to confirm the provider's current price page before making a decision. We will discuss the price logic in detail in the 6th unit.
OpenAI — GPT Family
OpenAI's GPT family is the most recognized brand in the market and has the largest third-party ecosystem; that is, many software tools, plug-ins and integrations are released for GPT first. The GPT family is also tiered: there are fast and cheap models (lightweight options like the GPT-4o mini, for example), balanced all-round models (GPT-4o), and thinking-heavy models like the "o-series" focused specifically on reasoning.
Areas where it stands out: Versatility and prevalence. Wide range of capabilities for text, code, images and audio; mature vehicle ecosystem; and broad community support. "Can a job be done with AI?" GPT is usually a ready starting point.
Typical user: Startups that do rapid prototyping, product teams that want broad integration, SMBs that want to handle a wide range of tasks with a single provider.
Google — Gemini Family
Google's Gemini family plays with two powerful cards: multimodality (the ability to handle text, images, audio, and video together) and Google Workspace integration (work embedded in tools like Gmail, Docs, Sheets). Gemini is also tiered: there are fast "Flash" models and more powerful "Pro" models.
Where it excels: Visual and video understanding, massive context, and frictionless integration for organizations that already use Google tools heavily. If an organization runs all office work on Google Workspace, it experiences Gemini as a natural extension.
Typical user: Companies established in the Google ecosystem, image/video-heavy content teams, those who want to process very large data sets at once.
Three Families Side by Side
Size
Anthropic (Claude)
OpenAI (GPT)
Google (Gemini)
strongest aspect
Long context, code, safe behavior
Versatility, broad ecosystem
Multimodality, Workspace
Stages
Opus/Sonnet/Haiku
Strong / balanced / light + reasoning
Pro/Flash
Typical user
Regulated sectors
Enterprise and product teams
Google focused institutions
Weight type
Closed
Closed
Closed
When looking at this table, do not ask "which one will win"; Ask "Which row does my job match?" If you are having a 200-page contract read, the long context line is decisive for you, and if you are having invoices read from a visual, the multimodality line is decisive for you.
Weak Prompt / Strong Prompt: When Asking for a Family Choice
Weak prompt:
Which is better: Claude, GPT or Gemini?
Powerful prompt:
Your role: neutral AI consultant. My job: I want to produce drafts to answer customer emails in an e-commerce company + extract amount/date from incoming invoice images. There are two different tasks: (1) text generation, (2) data extraction from the image. Task: Compare the Claude, GPT and Gemini families for both tasks. Write down which family stands out on a task basis, which tier I should use and why. Don't give an exact price, give a range. When in doubt, say "test".
Because the second prompt divides the job into two clear tasks, the model can recommend the right family for each task separately. This is also the basis for the "model portfolio" idea that we will learn about later.
Copiable Templates
Tier 1 matching:
I define the following task: [TASK]. Considering difficulty level and volume, which tier of the Claude, GPT and Gemini families (light/balanced/strong) would you recommend? Write one line of justification for each family.
2. Ecosystem compatibility:
Tools we currently use: [TOOL LIST, e.g. Google Workspace / Microsoft 365].Which AI family will fit most frictionlessly into this tool stack and why? Evaluate ease of integration, learning curve, and possible risk of lock-in (dependency).
3. Strengths validation:
I want to test the claim "[MODEL NAME] is the best at this" for my task [TASK]. Which 3 sample tests should I run at my own efficiency to verify this claim? Describe the success criteria for each test in one sentence.
4. Name resolution:
Parse the following model names into family, stage and version and make a table: [MODEL NAMES]. Label a typical use for each with a single word.
Common mistakes
- Deciding based on advertising: The announcement "This model came first in that test" may not say anything about your mission. Test it with your own data.
- Choosing a family without choosing a level: It is not enough to say "We will use GPT"; Which level it is changes radically the cost and quality.
- Ignoring ecosystem lock: Being deeply committed to a family brings ease of integration but makes it difficult to switch to another provider in the future.
- Missing the topic: This market changes fast; The comparison from six months ago may be invalid today. Base your decision on up-to-date information.
- Forcing one family to do everything: While one family shines in the text, another family may be ahead in the visual. Think task-by-task.
In summary
- Anthropic, OpenAI, and Google all offer the same tier logic (light/balanced/strong) under different names.
- Claude in long context, code and secure corporate behavior; GPT has great versatility and broad ecosystem; Gemini excels in multimodality and Workspace integration.
- “Best” varies by job; base the comparison on your own test data, not on advertising.
- Parsing model names into family + tier + version will help you find your way in the crowded market.
Application task
Identify two different tasks in your job (one text-heavy, one visual- or long-document-heavy). Using template 1 (tier matching) in this unit, have an AI model ask the appropriate rank of the three families for each task. Then, with template 3 (strengths validation), design three concrete tests to test the proposed model with your own data. We will use these tests in the evaluation study in unit 10.
checklist
- [ ] I can recognize the level (light/balanced/strong) logic in all three families.
- [ ] Claude, I can match the areas where GPT and Gemini are prominent.
- [ ] I can choose the right family and rank for a task with justification.
- [ ] I know ways to test a model claim with my own data.
- [ ] I can parse model names as family + stage + version.