Flex and Priority Service Tiers Now Available for Telnyx Inference

Telnyx Inference now offers Flex and Priority service tiers. Set service_tier on each request to trade latency against token cost, on the same models and the same synchronous API you already use.

What's new

• Priority on Kimi K2.6: Serving capacity optimized for interactive applications where lower latency matters more than the lowest token price. For teams building their own voice orchestration and other low latency use cases over Telnyx inference. Available in the US, EU, and UAE, with Australia coming next week.
• Flex on DeepSeek V4.1 Flash: Optimized for cost and throughput where end-to-end latency of several minutes is acceptable. Available in the US.
• Per-request selection: Set service_tier to flex, default, or priority on Chat Completions and Responses requests. Omitting the field uses default. Keep the same public model ID with no tier suffix, and no new endpoints are required.
• Same synchronous model: A tier changes serving capacity and rates while the request pattern stays the same. All requests remain synchronous, and Flex is not a batch API.
• Tier discovery via the models API: GET /v2/ai/openai/models lists data[].service_tiers per model. Not every model supports every tier, and the models API is the source of truth.

Why it matters

Interactive and background workloads compete for the same serving capacity, and until now price was the only lever. Service tiers make the tradeoff explicit: Priority keeps conversational turns fast on Kimi K2.6 for workloads where a late response breaks the experience, and Flex runs DeepSeek V4.1 Flash at lower rates for work where a few minutes of latency costs nothing. Both tiers run on Telnyx-hosted models, so you change one field in the request body rather than adding a second inference provider.

Example use cases

• Voice orchestration teams building their own agent loops on Kimi K2.6 Priority for interactive, latency-sensitive turns.
• Data and platform teams running offline evaluation, document processing, and background summarization on DeepSeek V4.1 Flash Flex.
• Cost-conscious teams routing each workload to the tier its latency requirement justifies, with tier support read from the models API.

Getting started

1. Check which models support each tier: call GET /v2/ai/openai/models and inspect data[].service_tiers.
2. Set service_tier in the request body:

curl https://api.telnyx.com/v2/ai/openai/chat/completions \
  -H "Authorization: Bearer $TELNYX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-ai/DeepSeek-V4.1-Flash",
    "service_tier": "flex",
    "messages": [
      {"role": "user", "content": "Summarize this support note in one sentence: The customer could not sign in. Resetting the password restored access."}
    ]
  }'

3. For Flex workloads, allow a longer request timeout and use bounded retries with exponential backoff. Honor Retry-After when present.

Learn more in the Inference service tiers docs or the Inference pricing page.