Preview models

New in preview: DeepSeek V4.1 Flash

DeepSeek V4.1 Flash (DeepSeek-V4.1-Flash-Preview) from DeepSeek is a general-purpose model with a 1M context window and up to 32K output tokens. It accepts text and image and returns text. On /chat/completions it supports streaming, function calling and reasoning: the model’s thinking comes back in reasoning_content. It is a preview model hosted on T-Cloud in Germany, within the EU, and, like the other preview models, free on any active plan. See Chat completions, Function calling and Reasoning.

New in preview: d3

d3 (d3-preview) is a 27B decision model from the open-weight Decision 3.0 family of vLLM Semantic Router. It is the recommended model for Decisions: /decisions returns typed answers (a probability, a choice or a score) instead of text, for fast agent steps such as tool routing, approval gates and grounding checks. It takes text and image input. d3 serves /decisions only; it does not generate text. See Decisions.

DiffusionGemma 26B A4B (diffusiongemma-26B-A4B-it-Preview), a diffusion-based Gemma model from Google with a 256K context window, serves both endpoints:

• Chat: /chat/completions with text and image input, streaming, function calling and an optional thinking mode. See Chat completions.
• Decisions: /decisions with text and image input. See Decisions.

Note

RPM and TPM values are marked with an asterisk ^* — click any asterisk for the conditions that apply.

Our Models

Get hands-on with the latest AI models: free to test, ready to explore. Our preview catalog is your sandbox to evaluate capabilities, benchmark performance, and prototype ideas before committing to production workloads. All you need is an active plan (Essential, Professional, or Agentic) to start.

Pick a model, try it in your workflow, and see what fits.

Disclaimer

Test models are provided free of charge with any active plan. Telekom may limit usage at any time — including available tokens, rate limits (TPM/RPM), and duration of use and may deactivate test models or restrict access without prior notice. Customers have no legal entitlement to the continued provision of test models.

ProviderAll

DeepSeekGoogle GemmavLLM

CloudAll

T-Cloud

LocationAll

Germany (EU)

All filtersFilters

All filters

Provider

DeepSeekGoogle GemmavLLM

Cloud

T-Cloud

Location

Germany (EU)

Input price

Up to €2€2 to €5Over €5

Output price

Up to €2€2 to €5Over €5

Clear allShow 3 models

Showing 3 of 3 models

Sort by Recommended orderModelProviderCloudLocationContext (tokens)Input (per 1M tokens)Output (per 1M tokens)Cached input (per 1M tokens) †Requests (per minute) *Tokens (per minute) *

Model	Provider	Cloud	Location	Context tokens	Input per 1M tokens	Output per 1M tokens	Cached input † per 1M tokens Price per million prompt tokens served from the prefix cache: 15% of the normal input price. T-Cloud-hosted language models only.	Requests * per minute This plan’s limit per minute. "dynamic": preview models share a quota that follows the load.	Tokens * per minute This plan’s limit per minute. "dynamic": preview models share a quota that follows the load.	Details
Google DiffusionGemma 26B A4BPreview	Google Gemma	T-Cloud Hosted on T-Cloud Public — Telekom's sovereign infrastructure in Germany, within the EU.	Germany (EU)	256K	€0.00	€0.00	€0.00	dynamic	dynamic	Show details: DiffusionGemma 26B A4B
vLLM d3Preview	vLLM	T-Cloud Hosted on T-Cloud Public — Telekom's sovereign infrastructure in Germany, within the EU.	Germany (EU)	—	€0.00	€0.00	€0.00	dynamic	dynamic	Show details: d3
DeepSeek DeepSeek V4.1 FlashNewPreview	DeepSeek	T-Cloud Hosted on T-Cloud Public — Telekom's sovereign infrastructure in Germany, within the EU.	Germany (EU)	1M	€0.00	€0.00	€0.00	dynamic	dynamic	Show details: DeepSeek V4.1 Flash

† Prompt tokens served from the prefix cache are billed at 15% of the normal input token price. This applies to T-Cloud-hosted language models only, not to embedding, reranking or speech-to-text models.

Worldwide means the third-party provider runs the model on its own infrastructure and may process data outside the EU/EEA‡. Enterprise Trust

‡ EU/EEA includes countries with an adequacy decision by the EU Commission.

Service note on Test / Preview models

^* Test and Preview models operate under dynamic rate limits rather than fixed Requests-per-Minute (RPM) or Tokens-per-Minute (TPM) ceilings. In addition:

• Telekom may adjust, restrict, or discontinue them at any time without prior notice.
• They carry no minimum-availability commitment and are not part of any Service Level Agreement (SLA).
• They are intended for evaluation and prototyping; they are not suitable for production workloads.
• Preview models are free on any active plan.

Customers requiring contractual guarantees on throughput, end-to-end latency, or time-to-first-token should consider Dedicated LLM Serving, where such commitments can be negotiated as performance SLAs on reserved hardware.