How to Choose SAI Models for Coding, Reasoning, and Large Codebases
A guide to picking the right Sistematis AI model: sai-1.0 through sai-1.4. Context windows, quota points, peak hours, and per-use-case recommendations — daily coding, deep reasoning, and 1M-token codebases.
Why Model Choice Matters#
Every request to the Sistematis AI API deducts quota points based on the model you pick. Choosing the right model for each task means your quota can last 3x longer — or you get far better reasoning quality exactly when you need it.
The rules are simple:
sai-1.0= 1 point per request- All other models (
sai-1.1–sai-1.4) = 3 points per request, 2 points off-peak (outside 13:00–17:00 WIB, Mon–Fri) - Final points scale with input/output tokens — requests with a large context cost more (Z.AI credit formula)
That means: with the LITE plan (135 points/5h), you get 135 sai-1.0 requests, or 67 flagship requests off-peak (0.5x rate), or 45 flagship requests during peak hours — as long as your context stays lean.
Sistematis AI Model Overview#
| Model | Positioning | Context | Max Output | Points/Request |
|---|---|---|---|---|
sai-1.4 | Flagship 1M | 1 million tokens | 131K tokens | 3 (2 off-peak) |
sai-1.3 | Deep Reasoning 1M | 1 million tokens | 131K tokens | 3 (2 off-peak) |
sai-1.2 | Professional | 200K tokens | 131K tokens | 3 (2 off-peak) |
sai-1.1 | Balanced | 200K tokens | 131K tokens | 3 (2 off-peak) |
sai-1.0 | Fast & Efficient | 200K tokens | 131K tokens | 1 (always) |
Money-saving tip: Quota points follow your input/output token volume (Z.AI credit formula). Short repeated prompts are very cheap, while dragging a huge context into every request burns points fast — start a new session when context bloats. Cache hits are much cheaper too. For light repetitive tasks,
sai-1.0is your best friend.
Recommendations per Use Case#
Everyday coding → sai-1.1 (Balanced)
Small refactors, new functions, bug fixes, generating unit tests. Low latency, stable quality, 3 points per request. This is the workhorse model for AI coding tools like Claude Code, Cline, and Kilo Code.
Simple tasks & quota saving → sai-1.0 (Fast)
Code formatting, autocomplete, file conversion, quick API questions. Fastest response and only 1 point. If you're close to running out of quota, switch to sai-1.0 to stay productive.
Deep analysis & debugging → sai-1.3 (Deep Reasoning 1M)
Hard-to-find bugs, architecture reviews, big migration planning, security analysis. Reasoning models think long before answering — slower per request, but they often find root causes faster models miss.
Large codebases & the most complex tasks → sai-1.4 (Flagship 1M)
A 1-million-token context means you can fit full documentation + dozens of files in one session. Use it for: understanding large monorepos, cross-module refactoring, or tasks that need the highest quality.
1M context note: on
sai-1.3andsai-1.4, the 1M context activates by appending the[1m]suffix to the model name in your tool config (e.g.sai-1.4[1m]) — only these two models truly support 1M.
Points-Saving Strategy#
Points burn fastest when context bloats — and off-peak (outside 13:00–17:00 WIB, Mon–Fri) the rate is just 0.5x. Three practical strategies:
- Start a new session when context bloats — accumulated input re-sent on every request is the biggest point drain
- Shift heavy tasks off-peak — code reviews or big analyses cost 50% less at night or on weekends
- Use
sai-1.0for light tasks — still 1 point at any time
The Sistematis AI dashboard always shows remaining quota in real time as percentages — check it before starting a big task.
Configuration Examples in AI Coding Tools#
Claude Code (Anthropic-compatible)
Claude Code automatically maps its model names to SAI tiers: Opus → sai-1.4, Sonnet → sai-1.1, Haiku → sai-1.0. You can also use SAI names directly in model settings. Full guide: Claude Code setup.
OpenAI SDK (Python)
from openai import OpenAI
client = OpenAI(
api_key="sai-xxxxxxxx",
base_url="https://api.sistematis.ai/v1",
)
response = client.chat.completions.create(
model="sai-1.3", # deep reasoning
messages=[{"role": "user", "content": "Review this service architecture..."}],
)
Codex, Cline, Kilo Code, Roo Code
All OpenAI-compatible tools just point at https://api.sistematis.ai/v1 with your Sistematis AI API key, then pick a sai-1.x model from the table above. See the tools documentation for per-tool guides.
Conclusion#
The most economical combination for most developers:
- Daily default:
sai-1.1— balance of quality and cost - Light tasks / saving:
sai-1.0— 1 point, fastest - Heavy artillery:
sai-1.3orsai-1.4— for analysis and large codebases
By matching the model to the task, even a LITE plan can power a full day of productivity.
Ready to try? Register free and feel the difference between tiers — or see the plans starting at Rp7,000/day.