01
Native by default
K-AI 1.0 appears first as the Kontinue-native option in the model picker.
The Kontinue AI model
Meet K-AI 1.0
The native intelligence layer inside Kontinue AI, designed to provide a complete AI experience while preserving the freedom to use other supported models.
What it is
Kontinue intelligence, presented transparently
We built K-AI 1.0 as an orchestration layer that works with underlying open-source models and Kontinue product capabilities such as memory, retrieval, projects, tasks, connectors, and specialised agents. We do not present it as a foundation model trained from scratch.
02
Product-aware
It works with Kontinue capabilities such as memory, retrieval, projects, tasks, connectors, and specialised agents.
03
Choice remains
Start with K-AI, import an existing conversation, or choose another supported model when the work calls for it.
Supported work
What you can do with K-AI 1.0
Everyday questionsWriting and rewritingBrainstormingSummarisationResearch assistanceReasoningCoding assistance
Why it exists
A native Kontinue experience, with model freedom
We believe Kontinue AI should offer more than a list of external APIs. K-AI brings our own product intelligence into the platform while we keep supported third-party models available for different tasks.
Runtime transparency
K-AI 1.0 at a glance
K-AI combines our product intelligence with an openly identified primary language model. These details describe the runtime currently configured in Kontinue AI.
Current version
1.0
Release date
1 June 2026
Primary language model
Gemma 4 31B Instruct
Supported languages
More than 140 languages through the primary Gemma 4 runtime
Published base-model results
Gemma 4 31B Instruct benchmarks
85.2%
MMLU Pro
89.2%
AIME 2026 · no tools
80.0%
LiveCodeBench v6
84.3%
GPQA Diamond
These are Google DeepMind's published results for the primary base model—not a claim that the complete K-AI product receives the same score. If the primary route is unavailable, K-AI can fall back to gpt-oss-120b.
Known limitations
- • Responses can be incomplete or incorrect and should be checked for important decisions.
- • Availability and underlying routing can change as the product evolves.
- • Published Gemma 4 results measure the base model, not the complete K-AI experience with Kontinue tools, memory, retrieval, and routing.
FAQ
Questions, answered
Does Kontinue AI have its own model?
K-AI 1.0 is our proprietary intelligence and orchestration layer, not a foundation model trained from scratch. Its primary language runtime is Gemma 4 31B Instruct, with gpt-oss-120b available as a reliability fallback.
Can K-AI 1.0 generate images?
The K-AI 1.0 language runtime produces text. When image generation is available in your plan, the Kontinue experience sends that work to our separate K-Image service; it is not generated by the K-AI language model itself.
Can I still use other AI models?
Yes. We also provide selected models from leading external providers. Model availability and limits depend on your plan.
Can I switch models during a conversation?
Yes. You can choose another supported model for the next response while keeping the conversation history in the same thread.
Can I import a previous conversation?
Yes. We support selected shared-link and export-file import methods. See the import guide for the current list.
Is K-AI 1.0 available on the free plan?
Yes. K-AI 1.0 is available on Free, Starter, Plus, Pro, and Max. Monthly request, context, output, and AI usage-credit allowances vary by plan.
Continue with Kontinue AI
Start with K-AI 1.0. Switch when the work needs something else.
One AI platform for native Kontinue intelligence, supported external models, and portable conversations.