> ## Documentation Index > Fetch the complete documentation index at: https://docs.mibyanai.com/llms.txt > Use this file to discover all available pages before exploring further. # Mibyan 4.1 > An Arabic-first model system, built on a K2.6 foundation and optimized for practical professional work Mibyan 4.1 is the public model system used across Mibyan and its developer API. It is built on a K2.6 foundation, then adapted for Arabic-first professional work: understanding the request behind the request, retaining constraints, and returning work that is ready to review—not merely fluent text. The public name is intentionally stable. The serving stack, routing policy, and runtime implementation can evolve without forcing application changes; the API contract and the behavioral bar remain the product. ## Public model ID Use this identifier in all new integrations: ```text theme={null} mibyan-4.1 ``` Some legacy aliases may be accepted during a migration window. Always prefer the identifier returned by `GET /v1/models`, as availability can depend on a project's access policy. ## Design target Mibyan is not presented as a general-purpose claim to win every possible benchmark. It is designed around a narrower, more useful question: *can a team reliably turn Arabic business intent into a usable first draft, structured result, or tool-assisted workflow?* | Area | Design target | What this looks like in practice | | ---------------------- | --------------------------------------------------------- | ----------------------------------------------------------------------------------------- | | Arabic fidelity | Preserve meaning, register, and terminology | A formal proposal stays formal; a Gulf business term is not flattened into generic MSA | | Mixed-language work | Understand code-switching as one instruction | Arabic requirements and English product names, schemas, or code can coexist in one prompt | | Constraint retention | Keep format, length, audience, and exclusions active | “Six bullets, for a CEO, no invented figures” is treated as a testable contract | | Professional structure | Prefer an editable working artifact over decorative prose | Plans include owners, decisions, risks, and next actions when the request calls for them | | Integration discipline | Behave predictably inside an application | Stable roles, streaming, structured-output pass-through, and tool-compatible message flow | ## What changed from the foundation K2.6 supplies the general language and reasoning base. Mibyan adds an adaptation layer focused on the behavior that Arabic professional workflows expose most quickly: * Modern Standard Arabic alongside Gulf terminology and regional business phrasing. * Arabic–English code switching, including product names, technical terms, tables, and JSON-shaped instructions. * Task examples for reports, proposals, meeting follow-ups, extraction, normalization, classification, and bilingual drafting. * Instruction-following examples that reward explicit structure, constraints, and safe uncertainty over generic completion. * Production-oriented message patterns: system, developer, user, assistant, and tool roles. This is an adaptation claim, not a claim that the model has perfect knowledge, unrestricted capability, or access to private customer data. ## Capabilities * Arabic and English text generation. * System, developer, user, assistant, and tool message roles. * Non-streaming and SSE streaming responses. * Tool and structured-output pass-through when the active model configuration supports them. * Chat Completions and the text/message subset of Responses. ## Where it fits best Choose Mibyan when Arabic quality and workflow discipline are part of the product requirement, not a cosmetic translation pass. Reports, proposals, executive summaries, policies, plans, and customer-facing drafts. Translation, terminology normalization, mixed Arabic–English requests, and cross-team handoffs. Extraction, classification, templates, and application-validated structured outputs. Agent flows where your application provides tools, validates results, and owns permissions. ## What Mibyan is—and is not Mibyan is a product model identity and API contract. Applications use `mibyan-4.1`; the Mibyan service owns the runtime configuration behind that public boundary. This separation lets us improve serving and safety without exposing infrastructure details or asking customers to rewrite requests. Mibyan is not currently documented as a general multimodal, embeddings, image, audio, or file API. Use the capabilities listed in the API reference as the source of truth. ## Recommended use cases * Arabic reports, proposals, business plans, and executive summaries. * Rewriting, translation, terminology normalization, and bilingual drafting. * Structured extraction and classification where the output schema is validated by your application. * Agent workflows where your application supplies tools and checks their results. For high-impact decisions, use retrieval or authoritative source material and require a review step. Benchmark results describe aggregate behavior; your own prompts are the final acceptance test. ## Disclosure boundaries We publish the model's public identity, K2.6-based training direction, intended behavior, evaluation protocol, API capabilities, and known limitations. We do not publish provider credentials, routing configuration, internal hosts, or a prompt that could weaken the service boundary. For the methodology behind these choices, read the [technical report](/models/technical-report), [evaluation protocol](/models/benchmarks), and [efficient serving note](/models/efficient-serving). ## API example ```python theme={null} from openai import OpenAI client = OpenAI( base_url="https://api.mibyanai.com/v1", api_key="mbn_test_...", ) response = client.chat.completions.create( model="mibyan-4.1", messages=[ {"role": "user", "content": "اكتب خطة إطلاق مختصرة لخدمة في مسقط."} ], ) print(response.choices[0].message.content) ```