UpgradeMajor upgrade underway: a new video studio, with more models coming soon. Stay tuned.

Multimodal open models, one API

Every model guaranteed genuine. One API key, pay as you go.

Already have an account?Sign in

POST /v1/chat/completionsbase_urlhttps://api.inferway.ai/v1
01Model & pricing

Chat & code

11

Coding, agents and everyday chat start here.

Flagshipmost capable
Text · Chatanthropic/claude-fable-5.2
Context1M
Max output128K
Usage limits
Text · Chatanthropic/claude-opus-5.5
Context1M
Max output128K
Usage limits
Text · Chatanthropic/claude-sonnet-5.5
Context1M
Max output128K
Usage limits
Text · Chatanthropic/claude-haiku-5.5
Context1M
Max output128K
Usage limits
Text · Chatmoonshotai/kimi-k3.1-pro
Context1M
Max output128K
Usage limits
Text · Chatz-ai/glm-5.3
Context1M
Max output128K
Zero data retentionUsage limits
Best valuecheap and fast
Text · Chatdeepseek/deepseek-v4.1-flash-1017-zdr
Context1M
Max output128K
Zero data retentionUsage limits
Text · Chatminimax/minimax-m3.1-flash
Context1M
Max output512K
FP8Usage limits
Text · Chatz-ai/glm-5.3-flash
Context1M
Max output131K
Zero data retentionUsage limits
Text · Chatxiaomi/mimo-v2.6-flash
Context1M
Max output128K
Usage limits

Uncensored

4

Versions with refusals removed, for fiction, role-play and security research.

Text · Chatinferway/qwen3.8-27b-uncensored
Context256K
Free tier1.5M tokens/day
Self-hostedNVFP4Zero data retention
Text · Chatz-ai/glm-5.3-uncensored
Context1M
Max output128K
Usage limits

Image

1

Text to image.

Image · Text-to-imageinferway/image-2.1-uncensored
Standard sizes5
Output formatPNG
Self-hostedBF16

Video

3

Text to video clips.

Video · Text to video
  • 480p
  • 720p
  • 1080p
Video · Text to video
  • 768p
  • 2K

Decision models

1

Typed answers for decision points: classification, choices, scores and image boxes, with probabilities.

Live
typesafe/jev-1.14
Input—
Output—

Ask typed questions about text or images and get structured answers with probabilities — no JSON parsing or retry loop in your app.

Endpoint: POST /v1/systemone

02Playground

Try it without an account.

Open full Console Playground
No guest model is available right now

Free trial — requests are not counted in your account.

Checking live model availability…

03Privacy

Private by construction.

Self-hosted ordinary text prompts and completions stay in memory and never touch disk. Request metadata is kept for up to 90 days.

01

Request sent

TLS-encrypted, straight to a dedicated endpoint.

02

Routed

No shared queue in the path.

03

Inference in memory

Paged KV, never written to disk.

04

Self-hosted text stays in memory

Optional MiniMax enhancement and future 2K are exceptions; H3 ciphertext retention and details are in the Privacy Policy.

04Integrate

Keep your SDK.

Create an accountSign up
no credit card
Point base_url at Inferway
https://api.inferway.ai/v1
Run the streaming smoke test
stream=true
Streaming
curl https://api.inferway.ai/v1/chat/completions \
  -H "Authorization: Bearer $INFERWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "inferway/qwen3.8-27b-uncensored",
    "stream": true,
    "messages": [{"role": "user", "content": "Say hello in one short sentence."}]
  }'
Copy setup for your AI assistant
Paste into Claude Code or Cursor. Key stays in your .env, never in the prompt.

Questions

Ordinary self-hosted text prompts and completions stay in memory without disk writes. Media results are kept for 7 days and uploads for up to 48 hours. H3 prompts, rewrites and MiMo drafts are encrypted and erased from the database within 7 days; encrypted backups rotate within 30 days, outside database erasure. Optional MiniMax enhancement and future 2K have separate disclosures in the Privacy Policy. Request metadata is kept for up to 90 days for billing and reliability.
Input tokens that hit the cache are billed at the cache rate rather than the normal input rate, so long shared system prompts and retrieval contexts benefit most. All three rates are the ones shown above, published for the market you are being served. The cache matches whole blocks of tokens, so a short prompt may not hit at all and then bills entirely at the input rate.
Recommended, not required. With stream=true, tokens arrive as soon as the first one is generated; without it, the whole answer arrives at the end.
You pay per token, metered from request metadata, never from your request content. Pricing and context limits are published on the model catalog and generated from a single catalog source.

Docs · Privacy · Transparency