Skip to main content
Model guide · October 2026

The right model
for every use case.

28 business use cases and the ccsio.ai model that solves each one, plus the model it works best with. Ten open-weight models, one API, running in the region you choose.

Two questions, one model

What do you need to do?

Choose the task and your main constraint. The recommendation updates instantly; the full list follows below.

1 · Type of task

2 · Main constraint

RecommendedComing soon

GLM 5.3

When complexity demands more than an answer.

✓ The strongest model in our catalog for this task.

Coming soon. Until it launches, the best option available today is Qwen 3.8.

Autonomous coding at near-frontier quality: open weights, strong on Terminal Bench 3.0 and long-horizon agent tasks.

Model details 1M ctx · 744B MoE · 40B active
Every use case

Each use case, its model.

Filter by area or tap a model to see only its use cases.

10 models cover all 28 use cases. Tap one to filter.

ChatReasoningDecisionsImagesTranscriptionEmbeddingsSpeech

28 use cases

Chat10

  1. 01 · ContentChat

    Summarize long email threads

    RecommendedGemma 4

    High volume, low complexity: a 4B-active model keeps the bill flat, and 256K of context holds the whole thread.

    Works best with: Qwen 3.8

  2. 02 · ContentChat

    Draft business content in several languages

    RecommendedGemma 4

    Everyday drafting for the whole company: efficient, multilingual and included in every plan.

    Works best with: Qwen 3.8

  3. 03 · ContentChat

    Marketing copy and A/B variants

    RecommendedGemma 4

    Dozens of variants per campaign at no extra token cost; FLUX.2 [klein] adds the visual for each one.

    Works best with: FLUX.2 [klein]

  4. 04 · ContentChat

    Policies and long internal documents

    RecommendedQwen 3.8

    Consistent terminology across hundreds of pages, because the full document fits in 1M tokens of context.

    Works best with: GLM 5.3

  5. 07 · KnowledgeChat

    Ask questions over a whole contract or manual

    RecommendedQwen 3.8

    No chunking needed: 1M tokens of context read the full document in a single pass.

    Works best with: Qwen3 Embedding 0.6B

  6. 09 · CodeChat

    Coding assistant for everyday development

    RecommendedQwen 3.8

    Included in every plan and able to hold an entire repository in context.

    Works best with: GLM 5.3

  7. 12 · CodeChat

    SQL from natural language

    RecommendedQwen 3.8

    Text-to-SQL over your own schema, without the schema ever leaving your region.

    Works best with: Clef

  8. 16 · OperationsChat

    Turn emails and chats into tasks

    RecommendedGemma 4

    Extracts who, what and when as JSON across everything your team writes in a day, at the lowest cost per message.

    Works best with: Clef

  9. 18 · CustomersChat

    Support assistant on your knowledge base

    RecommendedQwen 3.8

    Embeddings retrieve the right article; Qwen 3.8 answers in your customer’s language.

    Works best with: Qwen3 Embedding 0.6B

  10. 28 · AnalysisChat

    Survey and NPS analysis

    RecommendedGemma 4

    Themes and sentiment across thousands of open answers; embeddings group what customers say.

    Works best with: mxbai-embed-large v1

Reasoning4

  1. 10 · CodeReasoning

    Autonomous coding agents and long refactors

    RecommendedGLM 5.3Coming soon

    The strongest reasoner in our catalog for multi-step agentic work. Until it launches, Qwen 3.8 covers it.

    Works best with: Qwen 3.8

  2. 11 · CodeReasoning

    Code review across a large repository

    RecommendedDeepSeek 4.1 FlashComing soon

    Reasons over the full diff and its surrounding code with 1M tokens of context.

    Works best with: Qwen 3.8

  3. 26 · AnalysisReasoning

    Complex quantitative and logical analysis

    RecommendedGLM 5.3Coming soon

    Built for mathematical and multi-step reasoning, where a fast answer is not enough.

    Works best with: DeepSeek 4.1 Flash

  4. 27 · AnalysisReasoning

    Due diligence across a data room

    RecommendedDeepSeek 4.1 FlashComing soon

    Connects findings across thousands of pages with 1M tokens of context, inside the region you choose.

    Works best with: Qwen3 Embedding 0.6B

Decisions4

  1. 13 · OperationsDecisions

    Invoice approvals

    RecommendedClef

    Approve, reject or escalate with a confidence score for each option, so people review only the unclear cases.

    Works best with: Gemma 4

  2. 14 · OperationsDecisions

    Ticket routing

    RecommendedClef

    Picks the right team from your defined options and shows how sure it is.

    Works best with: mxbai-embed-large v1

  3. 15 · OperationsDecisions

    Classify scanned documents and photos

    RecommendedClef

    Text, images and video evaluated in one pass, ending in a structured decision instead of a paragraph.

    Works best with: Qwen 3.8

  4. 17 · OperationsDecisions

    Content moderation across text, images and video

    RecommendedClef

    Your own categories, a confidence score for each, and a clear threshold for human review.

    Works best with: Gemma 4

Images2

  1. 24 · VisualImages

    Product mockups and campaign visuals

    RecommendedFLUX.2 [klein]

    From a short brief to production-ready images in seconds; Gemma 4 writes the prompts at scale.

    Works best with: Gemma 4

  2. 25 · VisualImages

    Edit images with multiple references

    RecommendedFLUX.2 [klein]

    Keep the product, change the scene: several references guide one edit; Clef checks it against your brand rules.

    Works best with: Clef

Transcription3

  1. 20 · CustomersTranscription

    Call transcription and quality review

    RecommendedWhisper large v3

    Every call becomes searchable text; Qwen 3.8 scores it against your checklist.

    Works best with: Qwen 3.8

  2. 21 · VoiceTranscription

    Meeting minutes

    RecommendedWhisper large v3

    Whisper transcribes the recording, Qwen 3.8 writes the minutes and the action items.

    Works best with: Qwen 3.8

  3. 22 · VoiceTranscription

    Subtitles for videos and trainings

    RecommendedWhisper large v3

    Timestamped transcripts in nearly 100 languages; Gemma 4 adapts them for other markets.

    Works best with: Gemma 4

Embeddings3

  1. 05 · KnowledgeEmbeddings

    Semantic search over English documentation

    Recommendedmxbai-embed-large v1

    335M parameters and 1024-dimension vectors: precise retrieval for English content at a low cost per document.

    Works best with: Qwen 3.8

  2. 06 · KnowledgeEmbeddings

    Multilingual RAG across subsidiaries

    RecommendedQwen3 Embedding 0.6B

    One index for 100+ languages and code, so a question in Spanish finds the answer written in German.

    Works best with: Qwen 3.8

  3. 08 · KnowledgeEmbeddings

    Deduplicate and cluster a knowledge base

    Recommendedmxbai-embed-large v1

    Similar vectors reveal duplicates and topics; Clef then decides which version to keep.

    Works best with: Clef

Speech2

  1. 19 · CustomersSpeech

    Voice assistant for customer service

    RecommendedKokoro

    Whisper listens, a chat model answers and Kokoro speaks: 54 voices in 8 languages with low latency.

    Works best with: Whisper large v3

  2. 23 · VoiceSpeech

    Narration for e-learning and accessibility

    RecommendedKokoro

    Natural voices from an 82M-parameter model, efficient enough to voice a whole course catalog.

    Works best with: Gemma 4

How we choose

We recommend the model that fits the task and the cost profile of the workload, using the facts we publish for each model: context, size, capability and availability. Coming-soon models are marked and always paired with an option you can use today. This guide only covers models that run on ccsio.ai.

Explore the platform

Start with control

Create your account and pick the region your data must stay in.

Create accountTalk to us