Last week, we announced a new free “stealth” model inside Kilo Code called Spectre. At the time, we could only say this:
“A new stealth model from one of the top 10 AI model labs in the world.”
Today, we can finally tell you what Spectre actually is.
Spectre is Devstral 2, a next-generation agentic coding model from **Mistral AI. **Mistral AI has also launched its smaller sibling today, Devstral Small.
Here is the good news: Devstral 2 and Devstral Small will be free for all of December on Kilo Code.
Here’s what you can expect from the model.
Devstral 2 scored 72.2% on SWE-Bench Verified, reaching near parity with DeepSeek-V3.2 (73.1%) at a fraction of…
Last week, we announced a new free “stealth” model inside Kilo Code called Spectre. At the time, we could only say this:
“A new stealth model from one of the top 10 AI model labs in the world.”
Today, we can finally tell you what Spectre actually is.
Spectre is Devstral 2, a next-generation agentic coding model from **Mistral AI. **Mistral AI has also launched its smaller sibling today, Devstral Small.
Here is the good news: Devstral 2 and Devstral Small will be free for all of December on Kilo Code.
Here’s what you can expect from the model.
Devstral 2 scored 72.2% on SWE-Bench Verified, reaching near parity with DeepSeek-V3.2 (73.1%) at a fraction of the cost.
Devstral is also an **open-weights model, released under a modified MIT license. **The model is a dense 123B-parameter LLM, with a 256k context window.
All of this enables the model to achieve:
Deep code understanding at scale
Full agentic automation across large codebases
Efficient problem-solving with unmatched cost-performance
Architecture-level awareness enabled by the 256k context window
In comparison, Devstrall Small has 24B parameters, is locally deployable, and supports image inputs. The model scored 65.8% on SWE-bench Verified, and is competitive with models up to 5× its size.
We’ve had thousands of Kilo users test Spectre. Here are some of the things that stood out about this model to them:
Many noted that Spectre is very strong at following instructions.
Spectre showed high compliance and clarity when responding to guided prompts.
Many users were surprised at how fast the model is
Overall, our users see Spectre as promising, especially when properly set up and guided. Some used a frontier model in architect mode + had Spectre execute the instructions.
Devstral 2 costs $0.40 per 1M input tokens and $2.00 per 1M output tokens.
Devstral Small costs $0.10 per 1M input tokens, $0.30 per 1M output tokens.
**Devstral 2 and Devstral Small will stay free on Kilo Code through December. **
Try Devstral 2 and Devstral Small for a few days and tell us your impressions here. We’ll choose the best answers and provide 10 people with $50 in Kilo credits.
No posts