Saba with Jev
Saba is a model from Mistral, released 2025-02-17. It costs $0.200 per million input tokens and $0.600 per million output tokens, reads up to 33K tokens of context, and writes up to 26,214 tokens in one reply. Through llm11, set model to mistralai/mistral-saba.
- Input price
- $0.200 / 1M
- Output price
- $0.600 / 1M
- Context window
- 33K tokens
- Max output
- 26K tokens
- Cached input
- $0.020 / 1M
- Released
- 2025-02-17
- Accepts
- text, file
Read from the provider's prompt cache
Read from the live catalogue and refreshed daily. At blended list price, 72% of the 332 priced models we can call cost more.
Using Saba with Jev
Jev is TypeSafe AI's System One model. It reads each request first and decides which model in your pool answers, with a calibrated confidence on that call. Saba does the answering when Jev sends it work, or when you name it yourself. Saba falls in the llm11-fast band, and the labs and limits it meets make it eligible for that pack. Packs keep a spread of eight models per band, and it is not one of the eight today, so it is only picked when you name it in a custom pool.
Jev is new, so we have not published results for Saba with and without it, and this page does not invent any. Every response comes with a receipt that names the model that answered and what it cost against your most expensive model, so you can measure the pairing on your own traffic.
What a request costs
List price arithmetic, the same sum a receipt does. llm11 adds nothing per request; the only fee is on buying credits.
| Request | Tokens in / out | One | A thousand |
|---|---|---|---|
| Short chat turn | 500 / 200 | $0.00022 | $0.220 |
| Question over retrieved documents | 4,000 / 500 | $0.00110 | $1.10 |
| Long document summary | 30,000 / 1,000 | $0.00660 | $6.60 |
Cheaper models to try
The catalogue has no benchmark score for Saba, so this is a price ordering. It does not claim these answer as well.
| Model | Input / 1M | Output / 1M | Context | Released |
|---|---|---|---|---|
| Qwen Qwen3 Coder Next | $0.120 | $0.800 | 262K | |
| Mistral Mistral Small 4 | $0.150 | $0.600 | 262K | |
| Qwen Qwen3 VL 30B A3B Instruct | $0.150 | $0.600 | 262K | |
| Cohere Command R (08-2024) | $0.150 | $0.600 | 128K |
Call it
Same OpenAI request shape. Naming the model pins it, so nothing is routed, and the answer is still checked.
python
from openai import OpenAI
client = OpenAI(base_url="https://www.llm11.com/v1", api_key="llm11_live_...")
res = client.chat.completions.create(
model="mistralai/mistral-saba",
messages=[{"role": "user", "content": "Hello"}],
)More from Mistral
| Model | Input / 1M | Output / 1M | Context | Released |
|---|---|---|---|---|
| Mistral Medium 3.5 | $1.50 | $7.50 | 262K | |
| Mistral Small 4 | $0.150 | $0.600 | 262K | |
| Devstral 2 2512 | $0.400 | $2.00 | 262K | |
| Ministral 3 14B 2512 | $0.200 | $0.200 | 262K | |
| Ministral 3 3B 2512 | $0.100 | $0.100 | 131K | |
| Ministral 3 8B 2512 | $0.150 | $0.150 | 262K |
Questions
- What is Saba with Jev?
- Jev is TypeSafe AI's System One model. It reads each request first and decides which model in your pool answers, with a calibrated confidence on that call. Saba does the answering when Jev sends it work, or when you name it yourself. Saba falls in the llm11-fast band, and the labs and limits it meets make it eligible for that pack. Packs keep a spread of eight models per band, and it is not one of the eight today, so it is only picked when you name it in a custom pool. Routing decisions are made by Jev, TypeSafe AI's System One model.
- Is Saba a System One model?
- No. Saba answers requests. The System One model is Jev, which decides which model answers each one, so the two do different jobs and llm11 uses both.
- How much does Saba cost?
- $0.200 per million input tokens and $0.600 per million output tokens. A 4,000 token prompt with a 500 token answer costs about $0.00110, so a thousand of them is about $1.10. llm11 passes provider list price through and charges 5% when you buy credits.
- What is the Saba context window?
- 32,768 tokens, with up to 26,214 tokens in a single reply.
- Does Saba support tool calling and structured output?
- The catalogue lists tool calling and structured output, but it does not list reasoning controls.
- Can I use Saba through llm11?
- Yes. Set model to mistralai/mistral-saba on the OpenAI-compatible endpoint and the request goes to it directly, with verification still running.
- Does Jev route to Saba?
- Routing decisions are made by Jev, TypeSafe AI's System One model. Saba falls in the llm11-fast band, and the labs and limits it meets make it eligible for that pack. Packs keep a spread of eight models per band, and it is not one of the eight today, so it is only picked when you name it in a custom pool.
- What is a cheaper alternative to Saba?
- Nearest in price and below it: Qwen Qwen3 Coder Next at $0.120 in and $0.800 out, Mistral Mistral Small 4 at $0.150 in and $0.600 out and Qwen Qwen3 VL 30B A3B Instruct at $0.150 in and $0.600 out. The catalogue has no benchmark score for Saba, so this is a price ordering. It does not claim these answer as well.