When people try to make AI cheaper, their first instinct is usually "use it less." On a flat monthly subscription, the bigger savings usually come from "same task, which model does it." This article explains how Essevin key purposes work, how points are deducted, and how to dial in a point-saving setup that fits your usage pattern.
What Is a Key Purpose?
When you create a key on the console's API keys page, you choose under Use with which family of models it serves. Each key serves one purpose, and every key draws from the same plan points:
| Purpose | Serves | Typical client |
|---|---|---|
| Claude | Claude models | Claude Code, Anthropic SDK (address https://api.essevin.com, no /v1) |
| GPT | GPT chat and Codex | Codex CLI, OpenAI SDK, third-party clients (address https://api.essevin.com/v1) |
| Images | GPT Image models | OpenAI Images endpoints |
| Video models | Seedance and other video models (Everyday, Advanced and Professional plans) | Video jobs submitted through the API |
If you need two families, create one key for each — say, a Claude key for Claude Code and a GPT key for Codex. If you only chat, make images or make video in the browser, use web chat and Creative Studio directly; no key is needed. Using the wrong purpose is the most common setup mistake: calling an OpenAI-protocol endpoint with a Claude key, or running Claude Code with a GPT key, both fail.
What One Points Pool Means
Every key, web chat and Creative Studio share the same monthly points, at 10,000 points = HK$1. Each request is converted at "official price × credit rate" and deducted by actual usage, so point usage varies widely across models and tasks. The estimates published on the site (a standard message is 3,000 input and 800 output tokens):
| Use | Approx. points each |
|---|---|
| GPT-6 Sol message | ≈ 322 |
| Claude Opus 5.5 message | ≈ 1,771 |
| Codex coding request | ≈ 1,713 |
| GPT Image 2 image (1024², medium quality) | ≈ 7,872 |
| GPT Image 2 image (1024², low quality) | ≈ 879 |
| Seedance 2.0 clip (480p, 4 seconds) | ≈ 48,992 |
Switching tiers within a purpose works too: Claude Sonnet 5.5 is about half the official price of Opus 5.5, and Haiku 5.5 is lower still; GPT-6 Luna is far below GPT-6 Sol. Points follow the official price, so the right tier stretches the same pool several times further. Points reset every month and do not carry over.
Setup Recommendations for Three Profiles
- Solo developer: One Claude key for Claude Code, one GPT key for Codex. Use the Sonnet tier for everyday coding and switch to Opus only for complex refactors or hard bugs (type
/modelin the session); hand simple batch scripts to Haiku or GPT-6 Luna. Create and name one key per device and per tool so the usage records show exactly where points go. - Content and e-commerce creators: If you only make images and video in the browser, use Creative Studio directly; no key is needed. Draft images at low quality to test composition, then render medium or high once you like them; test video motion and camera work at 480p and short durations before raising resolution and length.
- Document and data batch processing: Run a small batch on a cheaper tier first; if spot checks pass, run the full set. Reserve the stronger model for the final check that needs judgement. Put the fixed prefix of each batch (system prompt, examples) first to capture cache discounts.
Order of Point Optimization
First, capture cache discounts (prompt structure, zero cost). Second, pick the right model tier (configure once, benefit every day). Third, cut usage (a real tradeoff, move carefully). Follow that order and most people make their monthly points last noticeably longer without sacrificing quality — the usage records never lie. Subscriptions are now open; see each plan's points and included models at essevin.com and subscribe at console.essevin.com.