Enter passcode
Incorrect passcode
Ven Agency — Internal Resource

AI
Model
Guide

Every model we use, what it does best, and how to use it inside Telegram.

Scroll
Provider 01 of 03

Anthropic
Claude

Three models. Matched to task complexity.

01
Anthropic 🔒 Restricted — Request from Steve
Opus 4.6
Maximum intelligence. Use when quality matters more than speed.
🧠
Complex Strategy
Long-term plans, difficult decisions, nuanced analysis.
✍️
High-Stakes Writing
Pitch decks, proposals, important client communications.
🔍
Deep Research
Multi-step research where errors are costly.
⚠️
Watch out for
Overthinking. Opus can over-explain simple requests and be slower than needed. Don't use it for quick tasks — you'll wait longer for the same quality Sonnet would give you. Also expensive — avoid for bulk or repetitive use.
🔒
Restricted Model
Opus is not available for general use. If you believe your task genuinely requires it, message Steve on Telegram to request access. In most cases, Sonnet will do the job just as well.
Slower
$$$
/new opus — restricted
02
Anthropic
Sonnet 4.6
Your default. The best balance of smart and fast.
📧
Everyday Writing
Emails, Slack messages, content, documentation.
📊
Standard Analysis
CRM summaries, reports, data interpretation.
🗂️
Most Workflows
95% of daily tasks. If unsure, start here.
⚠️
Watch out for
Confident hallucinations. Sonnet will sometimes state incorrect facts with full confidence — especially dates, numbers, and URLs. Always verify specific data points it gives you. Never blindly trust a figure without checking.
Fast
$$
/new sonnet
03
Anthropic
Haiku 4.5
The fastest, cheapest Claude. For quick and repetitive tasks.
⚡
Quick Lookups
Single facts, quick summaries, basic answers.
🔁
Repetitive Tasks
Bulk classification, data labelling, simple transforms.
💬
Automations
Background agent steps where speed beats depth.
⚠️
Watch out for
Shallow reasoning. Haiku will give short, surface-level answers to complex questions — and won't tell you it's doing so. If the reply feels too simple or misses nuance, upgrade to Sonnet. Not suitable for anything strategic.
Instant
$
/new haiku
Provider 02 of 03

Google
Gemini

Native to the Google ecosystem. Best for Ads, Analytics, and large data.

04
Google
Gemini 3.1 Pro
The go-to for Google Ads, Analytics, and Search Console work.
📈
Google Ads
Campaign analysis, performance reports, keyword work.
📉
GA4 & Search Console
Traffic analysis, SEO insights, conversion tracking.
🖼️
Multimodal Tasks
Analysing screenshots, images alongside text.
⚠️
Watch out for
Weaker creative writing. Gemini's writing style can feel formulaic and stiff compared to Claude — don't use it for copy or client-facing content. Also occasionally verbose; you may need to prompt it to be more concise.
Fast
$$
/new gemini 3.1
05
Google
Gemini 2.5 Pro
For huge datasets, long documents, and deep data extraction.
📂
Large Spreadsheets
Thousands of rows. Massive exports and data sets.
📄
Long Documents
Contracts, lengthy reports, full-length research docs.
🔗
Cross-source Research
Combining multiple data sources into one analysis.
⚠️
Watch out for
Slow on simple tasks. The large context window is overkill for short tasks — it can feel sluggish compared to Flash. It also has a tendency to produce very long responses. Prompt with "be concise" if you're getting walls of text.
Fast
$$
/new gemini 2.5
06
Google
Gemini 3 Flash
High speed, high volume. Cheapest Google option.
🏎️
Bulk Operations
Classifying large batches of leads or content items.
🔄
Fast Fallback
When a task is simple and speed is the priority.
🤖
Agent Automation
Background tasks inside automated workflows.
⚠️
Watch out for
Low accuracy on nuanced tasks. Flash will confidently give wrong answers on anything complex. It's not suitable for strategy, writing, or anything where context matters. Speed comes at the cost of reasoning quality.
Instant
$
/new flash
Provider 03 of 03

OpenAI
GPT

One model. One job. Do not use for anything else.

07
OpenAI
GPT-5.3 Codex
Specialist coding model. Use exclusively for technical work.
💻
Writing Code
Scripts, automation, new features, full functions.
🐛
Debugging
Finding and fixing errors in existing code.
🔎
Code Review
Reviewing pull requests, auditing logic, refactoring.
⚠️
Watch out for
Terrible at non-code tasks. Do not use Codex for writing, strategy, or analysis — it's a specialist, not a generalist. It can also be overly literal when given vague briefs; always describe exactly what the code needs to do.
Fast
$$
/new codex
Using the agents

Switching Models
in Telegram

Two commands. That's it.

Start Fresh with a Model

Resets the conversation and starts with the model you pick.

You
/new sonnet
Agent
✅ Switched to Claude Sonnet 4.6. New session started.
🔒 /new opus is restricted. Message Steve on Telegram to request access.

Mid-Chat Switch

Changes model without clearing context.

You
/model gemini 3.1
Agent
✅ Model updated to Gemini 3.1 Pro.

See All Available Models

Opens a numbered list to pick from.

You
/model list
Agent
1. Sonnet 4.6 [Steve] ✅
2. Gemini 3.1 Pro
3. Opus 4.6 [Steve]
...
The big one

Rate Limits

The most common issue on the team. Here's what's happening and what to do.

What is a rate limit?

Each AI provider limits how many requests can be made per minute or per day. When the team is all using the same agent at the same time, or a single conversation runs too many back-and-forth messages quickly, the provider blocks further requests temporarily.

What does it look like?
Rate limit exceeded. Please try again in 60 seconds.
429: Too Many Requests
⚠️ You exceeded your current quota
What to do right now
  1. 1 Wait 60 seconds then retry. Most rate limits are per-minute and reset quickly.
  2. 2 Switch to a different provider. If Claude is rate-limited, switch to Gemini: /new gemini 3.1
  3. 3 Avoid back-to-back requests. Don't send 10 messages in a row — let the agent finish and respond before sending the next one.
  4. 4 If it persists for more than 10 minutes across all providers — message Steve. Could be a quota or billing issue.
💡 Pro tip: Rate limits are per-provider, not per-agent. If Anthropic (Claude) is rate-limited, Gemini and GPT are unaffected. Always try switching providers before giving up.
Self-serve fixes

Common Errors

Try these before escalating.

🔄 Agent Not Responding / Slow
Agent is thinking... (no reply after 2+ min)
Send /stop to abort, then /reset to clear the session. Usually fixes it instantly.
💭 Weird / Off-Topic Replies
Agent is giving irrelevant or confused answers
The chat context has become too long. Send /reset to start a clean session.
🔑 API / Billing Error
⚠️ API provider returned a billing error
The API key has hit its quota or expired. Send to Steve immediately — this can't be fixed from chat.
⛔ Command Not Working
/model sonnet → no response
Make sure the command is the only thing on that line. No text before or after. Send it as a standalone message.
📋 Hallucinated Data / Wrong Facts
Agent confidently stated something incorrect
Point it out: "That's wrong, [correct info]". It will acknowledge and correct. Always verify critical data independently.
⏳ Check Agent Status
Not sure if the agent is working?
Send /status for a health check. Shows current model, context usage, and whether the provider is healthy.
When to escalate

Can't Fix
It?
Send It.

If /reset doesn't solve it, don't spend more time on it. Copy the exact error message and send it to Steve on Telegram.

Try in this order first

  1. 1 Send /reset — clears 90% of issues.
  2. 2 Send /status — checks if provider is up.
  3. 3 Try switching models: /new sonnet or /new flash.
  4. 4 Still broken? Copy the error message and send directly to Steve. No explanation needed.
Send to: Steve on Telegram — paste the error, nothing else.
Questions or access requests
Message Steve on Telegram
@stee_ven · tap to open Telegram