Choosing an AI model

F68 currently offers:

GPT · CLAUDE · GEMINI · GROK · DEEPSEEK · GLM · KIMI · QWEN

There is no best model for every job.

Choose in three layers:

AI model family → Model/tier → Reasoning level

For example:

GPT → Terra → Medium

or:

Claude → Sonnet → High

Choose the right AI family

AI family	Strengths	Best for
GPT	Planning, workflows, following structure, knowledge base, report writing	Planning, analysis, Agents, multi-step workflows
Claude	Research, deep analysis, coding, Bots, reading documents	Research, coding, Agents, critique
Gemini	Large context, multimodal, documents and office work	PDFs, images, video, long documents, office
Grok	Strong general capability, competitive cost/performance	Research, coding, everyday professional work
DeepSeek	Reasoning, coding, low cost	Analysis, coding, high-volume workflows
GLM	Coding, Agents, automation	Bots, Agents, automated workflows
Kimi	Long context, research, coding, Agents	Large document collections and extended research
Qwen	General-purpose, coding, multilingual, competitive cost	Everyday work, automation, high-volume

This is guidance on how to use them, not an absolute ranking.

GPT

F68 currently offers:

GPT SOL

For hard and important problems.

Best for:

• strategy;
• in-depth analysis;
• complex planning;
• long workflows;
• complex coding tasks;
• agents used for high-stakes work;
• critique.

Prioritizes quality over cost.

GPT TERRA

A model that balances quality and cost.

Best for:

• analysis;
• research;
• planning;
• report writing;
• knowledge base;
• coding;
• Agents;
• professional workflows.

This is the GPT option to try first for most professional work.

GPT LUNA

Prioritizes speed and cost.

Best for:

• everyday chat;
• summarization;
• translation;
• rewriting;
• extraction;
• classification;
• data processing;
• high-volume tasks.

Things to note about GPT

GPT usually organizes work and follows workflows well, but it sometimes tends to:

• analyze more broadly than requested;
• suggest extras beyond the scope;
• overthink when the reasoning level is too high.

For narrow tasks, state clearly the scope to be covered.

Claude

F68 currently offers:

CLAUDE FABLE

Claude’s top model.

Best for:

• complex research;
• complex coding;
• system architecture;
• long-running Agents;
• in-depth analysis;
• problems where mistakes are costly.

CLAUDE OPUS

A premium option that balances capability and cost.

Best for:

• research;
• finance;
• coding;
• knowledge work;
• Agents;
• regular professional analysis.

CLAUDE SONNET

Claude’s workhorse model.

Best for:

• everyday research;
• coding;
• analysis;
• writing;
• reading documents;
• Agents;
• using tools.

If you like how Claude works but don’t need a flagship, start with Sonnet.

CLAUDE HAIKU

Prioritizes speed and cost.

Best for:

• chat;
• summarization;
• extraction;
• classification;
• small coding tasks;
• sub-agents;
• batch processing.

Things to note about Claude

Claude is strong at digging deep and coding, but in some long workflows it may shorten the process on its own or skip a step.

For processes that require many mandatory steps, use:

• numbered workflow;
• checklist;
• output schema;
• completion rule.

Gemini

F68 currently offers:

GEMINI PRO

Prioritizes quality.

Best for:

• long documents;
• PDFs;
• images;
• video;
• multimodal;
• research;
• document analysis;
• complex office work.

GEMINI FLASH

Prioritizes speed and cost.

Best for:

• document processing;
• summarization;
• extraction;
• office tasks;
• routine multimodal tasks;
• large workloads.

When to choose Gemini

If the problem involves:

many documents + many formats + large context

Gemini is usually an option to consider early.

Grok

Consider Grok when balancing capability and cost.

Best for:

• research;
• analysis;
• coding;
• knowledge work;
• everyday professional work.

You can use Grok first before moving to more expensive flagship models.

DeepSeek

DeepSeek stands out for:

• reasoning;
• coding;
• Agents;
• cost;
• handling large volumes.

Best for:

• analysis;
• coding;
• automation;
• preprocessing;
• multi-request workflows;
• everyday tasks.

Especially useful when cost per run is an important factor.

GLM

GLM focuses strongly on:

• coding;
• Agents;
• automation;
• tool use;
• long-horizon workflows.

Best for:

• building Bots;
• coding;
• Agents;
• automated workflows;
• sub-agents;
• tasks that run frequently.

Kimi

Kimi suits:

• very long context;
• research;
• many documents;
• coding;
• Agents;
• knowledge work.

Consider it when a task has to read and process a large amount of context before reaching a conclusion.

Qwen

Qwen is a general-purpose AI family with a cost/performance advantage.

Best for:

• chat;
• coding;
• translation;
• extraction;
• summarization;
• multilingual;
• Agents;
• high-volume workloads.

Don’t just choose a vendor — choose a tier

There are two ways to reduce cost.

Option 1 — Keep the vendor, lower the tier

GPT: Sol → Terra → Luna
Claude: Fable → Opus → Sonnet → Haiku
Gemini: Pro → Flash

For example:

A job that suits Claude doesn’t mean you have to use Fable.

Sonnet may be good enough for most everyday research and coding.

Likewise, a job that suits GPT can use Terra or Luna instead of Sol.

Option 2 — Choose an AI family with better cost/performance

You can consider:

Grok · DeepSeek · GLM · Kimi · Qwen

especially for frequent or high-volume workloads.

Quick picks by task

Job	Try first
Everyday chat	Luna / Haiku / Flash / DeepSeek / Qwen
Translation, summarization, extraction	Luna / Haiku / Flash / Qwen
Writing articles	Terra / Sonnet
Planning	Terra / Sol
Research	Sonnet / Opus / Terra / Gemini / Kimi
Very hard research	Fable / Sol / Opus
In-depth analysis	GPT / Claude
Coding	Sonnet / Opus / GPT / GLM
Building Bots / Agents	Claude / GPT / GLM
Multi-step workflows	GPT / Claude
Working with KB	GPT
Very long documents	Gemini / Kimi
PDFs / images / video	Gemini
Office work	Gemini
High-volume automation	Luna / Haiku / Flash / DeepSeek / GLM / Qwen
Cost optimization	Terra / Sonnet / Grok / DeepSeek / GLM / Kimi / Qwen
Very important work	Flagship + critique by another model