Gateway 101 priced models · 7 providers · 10 aliases

One endpoint for every model.

An OpenAI-compatible API in front of Anthropic Claude, Amazon Bedrock, OpenAI, DeepSeek, Google, Deepgram and ElevenLabs, plus TypeSafe's Jev for calibrated decisions. Every response carries its exact cost, every key has a hard budget, and every call is logged and fails over automatically.

Read the docs Browse the catalog

Change base_url, keep your SDK. Works with openai, Anthropic Messages and plain HTTP.

POST /v1/chat/completions
// request
{ "model": "fast",
  "messages": [{ "role": "user", "content": "Hello" }] }

// response (excerpt)
{ "model": "bedrock/us.amazon.nova-lite-v1:0",
  "cost": {
    "alias": "fast", "currency": "USD",
    "total_usd": 0.00000204,
    "budget": { "limit_usd": 20, "remaining_usd": 19.99998 } } }
101
Priced models, live
7
Providers behind one key
3
Replicas with failover
100%
Calls costed & logged

Why a gateway

Provider sprawl, solved once

Apps ask for smart or fast. The gateway picks the model, enforces the money rules and tells you what it cost.

One API, every provider

OpenAI Chat Completions, Embeddings, Audio and Anthropic Messages on the same host. Tools, vision, streaming and reasoning pass through.

Cost on every response

A cost object with input, output and total USD plus remaining budget, in the body and in x-models-cost-usd. Streams carry it in the final chunk.

Budgets that hold

Daily, monthly or total USD caps per key, enforced by reserving worst-case cost before the call. Per-end-user budgets from the user's plan.

Guardrails

RPM and TPM limits, max-token caps, model allowlists, alert thresholds to Slack and Prometheus, and a global monthly ceiling.

Failover before the first byte

Aliases are ordered chains. Throttling, 5xx, timeouts and unavailable models fall through to the next target; a 400 never does.

Full logging

Metadata for every call, always. Request and response bodies only when enabled per key, AES-256-GCM encrypted with 30-day retention.

Live catalog

Aliases, mapped by admins

Use an alias instead of a vendor id. When a better or cheaper model ships, the mapping changes and your code does not.

cheap

chat

The cheapest acceptable chat model, for very high volume.

Resolves to Amazon Nova Micro (us)

  1. Amazon Nova Micro (us)
  2. OpenAI gpt-oss-20b
  3. Amazon Nova Lite (us)

code

chat

Code generation, review and refactoring.

Resolves to Claude Sonnet 5.5

  1. Claude Sonnet 5.5
  2. Claude Sonnet 5
  3. Qwen Qwen3-Coder-30B-A3B-Instruct
  4. deepseek-flash

decision

decision

Calibrated typed decisions (noul / choice / score / box with probabilities and confidence) from TypeSafe Jev. Only valid on POST /v1/systemone — not a chat model.

Resolves to Jev 1.14

  1. Jev 1.14
  2. Jev 1.13

embed

embedding

Text embeddings, 1024 dimensions (Amazon Titan Text Embeddings v2). Single target: never fails over to a different vector space.

Resolves to Amazon Titan Text Embeddings V2

  1. Amazon Titan Text Embeddings V2

fast

chat

Low-latency, low-cost chat for short tasks, routing and extraction.

Resolves to Amazon Nova Lite (us)

  1. Amazon Nova Lite (us)
  2. deepseek-flash
  3. Qwen Qwen3 32B (dense)
  4. gpt-4.1-nano

reasoning

chat

Multi-step reasoning, planning and hard problems. Slower; returns reasoning_content when the model exposes it.

Resolves to Claude Opus 5.5

  1. Claude Opus 5.5
  2. Claude Opus 5
  3. gpt-5
  4. bedrock/global.anthropic.claude-opus-5
  5. Anthropic Claude Sonnet 4.5 (global)
  6. deepseek-v4-pro

smart

chat

Balanced quality for most product features: writing, analysis, tool use.

Resolves to Claude Sonnet 5.5

  1. Claude Sonnet 5.5
  2. Claude Sonnet 5
  3. bedrock/global.anthropic.claude-sonnet-5
  4. Anthropic Claude Haiku 4.5 (us)
  5. gpt-4.1-mini
  6. deepseek-v4-pro

transcribe

transcription

Speech-to-text for recorded audio files (multilingual).

Resolves to Deepgram Nova-3

  1. Deepgram Nova-3
  2. gpt-4o-mini-transcribe
  3. whisper-1

tts

speech

Premium text-to-speech and cloned voices (ElevenLabs). voice = an ElevenLabs voice id; an ElevenLabs voice never fails over to another vendor.

Resolves to Eleven Multilingual v2

  1. Eleven Multilingual v2
  2. Eleven Flash v2.5
  3. tts-1

vision

chat

Image understanding (send image_url parts as base64 data URLs).

Resolves to Claude Haiku 4.5

  1. Claude Haiku 4.5
  2. Anthropic Claude Haiku 4.5 (us)
  3. gpt-4.1-mini
  4. Amazon Nova Lite (us)

All models

Prices in USD per 1M tokens unless noted. Only enabled and priced models are listed.

ModelProviderTypeInputOutput
Claude Fable 5anthropic/claude-fable-5 anthropic chat $10.00 $50.00
Claude Fable 5.1anthropic/claude-fable-5-1 anthropic chat $10.00 $50.00
Claude Haiku 4.5anthropic/claude-haiku-4-5-20251001 anthropic chat $1.00 $5.00
Claude Haiku 5.5anthropic/claude-haiku-5-5 anthropic chat $0.10 $0.50
Claude Opus 4.5anthropic/claude-opus-4-5-20251101 anthropic chat $5.00 $25.00
Claude Opus 4.6anthropic/claude-opus-4-6 anthropic chat $5.00 $25.00
Claude Opus 4.7anthropic/claude-opus-4-7 anthropic chat $5.00 $25.00
Claude Opus 4.8anthropic/claude-opus-4-8 anthropic chat $5.00 $25.00
Claude Opus 5anthropic/claude-opus-5 anthropic chat $5.00 $25.00
Claude Opus 5.5anthropic/claude-opus-5-5 anthropic chat $4.00 $20.00
Claude Sonnet 4.6anthropic/claude-sonnet-4-6 anthropic chat $3.00 $15.00
Claude Sonnet 5anthropic/claude-sonnet-5 anthropic chat $2.00 $10.00
Claude Sonnet 5.5anthropic/claude-sonnet-5-5 anthropic chat $2.00 $10.00
Amazon Nova Litebedrock/amazon.nova-lite-v1:0 bedrock chat $0.06 $0.24
Amazon Nova Microbedrock/amazon.nova-micro-v1:0 bedrock chat $0.035 $0.14
Amazon Nova Probedrock/amazon.nova-pro-v1:0 bedrock chat $0.80 $3.20
Amazon Titan Embeddings G1 - Textbedrock/amazon.titan-embed-text-v1 bedrock embedding $0.10 —
Amazon Titan Text Embeddings V2bedrock/amazon.titan-embed-text-v2:0 bedrock embedding $0.02 —
Amazon Nova 2 Lite (global)bedrock/global.amazon.nova-2-lite-v1:0 bedrock chat $0.30 $2.50
Anthropic Claude Haiku 4.5 (global)bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0 bedrock chat $1.00 $5.00
Anthropic Claude Sonnet 4.5 (global)bedrock/global.anthropic.claude-sonnet-4-5-20250929-v1:0 bedrock chat $3.00 $15.00
OpenAI gpt-oss-120bbedrock/openai.gpt-oss-120b-1:0 bedrock chat $0.15 $0.60
OpenAI gpt-oss-20bbedrock/openai.gpt-oss-20b-1:0 bedrock chat $0.07 $0.30
Qwen Qwen3 32B (dense)bedrock/qwen.qwen3-32b-v1:0 bedrock chat $0.20 $0.78
Qwen Qwen3-Coder-30B-A3B-Instructbedrock/qwen.qwen3-coder-30b-a3b-v1:0 bedrock chat $0.15 $0.60
Qwen Qwen3 Coder Nextbedrock/qwen.qwen3-coder-next bedrock chat $0.50 $1.20
Qwen Qwen3 Next 80B A3Bbedrock/qwen.qwen3-next-80b-a3b bedrock chat $0.15 $1.20
Qwen Qwen3 VL 235B A22Bbedrock/qwen.qwen3-vl-235b-a22b bedrock chat $0.53 $2.66
Amazon Nova 2 Lite (us)bedrock/us.amazon.nova-2-lite-v1:0 bedrock chat $0.30 $2.50
Amazon Nova Lite (us)bedrock/us.amazon.nova-lite-v1:0 bedrock chat $0.06 $0.24
Amazon Nova Micro (us)bedrock/us.amazon.nova-micro-v1:0 bedrock chat $0.035 $0.14
Amazon Nova Pro (us)bedrock/us.amazon.nova-pro-v1:0 bedrock chat $0.80 $3.20
Anthropic Claude Haiku 4.5 (us)bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0 bedrock chat $1.10 $5.50
Anthropic Claude Sonnet 4.5 (us)bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0 bedrock chat $3.30 $16.50
DeepSeek DeepSeek-R1 (us)bedrock/us.deepseek.r1-v1:0 bedrock chat $1.35 $5.40
Meta Llama 3.1 70B Instruct (us)bedrock/us.meta.llama3-1-70b-instruct-v1:0 bedrock chat $0.72 $0.72
Meta Llama 3.1 8B Instruct (us)bedrock/us.meta.llama3-1-8b-instruct-v1:0 bedrock chat $0.22 $0.22
Meta Llama 3.3 70B Instruct (us)bedrock/us.meta.llama3-3-70b-instruct-v1:0 bedrock chat $0.72 $0.72
Meta Llama 4 Maverick 17B Instruct (us)bedrock/us.meta.llama4-maverick-17b-instruct-v1:0 bedrock chat $0.24 $0.97
Meta Llama 4 Scout 17B Instruct (us)bedrock/us.meta.llama4-scout-17b-instruct-v1:0 bedrock chat $0.17 $0.66
Mistral AI Pixtral Large (25.02) (us)bedrock/us.mistral.pixtral-large-2502-v1:0 bedrock chat $2.00 $6.00
Deepgram Nova-2deepgram/nova-2 deepgram transcription $0.0043 /min —
Deepgram Nova-3deepgram/nova-3 deepgram transcription $0.0052 /min —
deepseek-flashdeepseek/deepseek-flash deepseek chat $0.22 $0.66
deepseek-v4-prodeepseek/deepseek-v4-pro deepseek chat $0.66 $1.98
Eleven Flash v2elevenlabs/eleven_flash_v2 elevenlabs speech $40.00 /1M chars —
Eleven Flash v2.5elevenlabs/eleven_flash_v2_5 elevenlabs speech $40.00 /1M chars —
Eleven Multilingual v2elevenlabs/eleven_multilingual_v2 elevenlabs speech $80.00 /1M chars —
Eleven Turbo v2elevenlabs/eleven_turbo_v2 elevenlabs speech $40.00 /1M chars —
Eleven Turbo v2.5elevenlabs/eleven_turbo_v2_5 elevenlabs speech $40.00 /1M chars —
Eleven v3elevenlabs/eleven_v3 elevenlabs speech $80.00 /1M chars —
Eleven v3 Conversationalelevenlabs/eleven_v3_conversational elevenlabs speech $40.00 /1M chars —
Eleven v4elevenlabs/eleven_v4 elevenlabs speech $80.00 /1M chars —
Eleven v4 Turboelevenlabs/eleven_v4_turbo elevenlabs speech $40.00 /1M chars —
Jev 1.13jev/jev-1.13.0 jev decision $0.042 $0.00
Jev 1.14jev/jev-1.14.0 jev decision $0.042 $0.00
jev-latestjev/jev-latest jev decision $0.042 $0.00
jev-previewjev/jev-preview jev decision $0.042 $0.00
gpt-4.1openai/gpt-4.1 openai chat $2.00 $8.00
gpt-4.1-2025-04-14openai/gpt-4.1-2025-04-14 openai chat $2.00 $8.00
gpt-4.1-miniopenai/gpt-4.1-mini openai chat $0.40 $1.60
gpt-4.1-mini-2025-04-14openai/gpt-4.1-mini-2025-04-14 openai chat $0.40 $1.60
gpt-4.1-nanoopenai/gpt-4.1-nano openai chat $0.10 $0.40
gpt-4.1-nano-2025-04-14openai/gpt-4.1-nano-2025-04-14 openai chat $0.10 $0.40
gpt-4oopenai/gpt-4o openai chat $2.50 $10.00
gpt-4o-2024-05-13openai/gpt-4o-2024-05-13 openai chat $2.50 $10.00
gpt-4o-2024-08-06openai/gpt-4o-2024-08-06 openai chat $2.50 $10.00
gpt-4o-2024-11-20openai/gpt-4o-2024-11-20 openai chat $2.50 $10.00
gpt-4o-miniopenai/gpt-4o-mini openai chat $0.15 $0.60
gpt-4o-mini-2024-07-18openai/gpt-4o-mini-2024-07-18 openai chat $0.15 $0.60
gpt-4o-mini-transcribeopenai/gpt-4o-mini-transcribe openai transcription $0.003 /min —
gpt-4o-mini-transcribe-2025-03-20openai/gpt-4o-mini-transcribe-2025-03-20 openai transcription $0.003 /min —
gpt-4o-mini-transcribe-2025-12-15openai/gpt-4o-mini-transcribe-2025-12-15 openai transcription $0.003 /min —
gpt-4o-transcribeopenai/gpt-4o-transcribe openai transcription $0.006 /min —
gpt-5openai/gpt-5 openai chat $1.25 $10.00
gpt-5-2025-08-07openai/gpt-5-2025-08-07 openai chat $1.25 $10.00
gpt-5.6-lunaopenai/gpt-5.6-luna openai chat $0.22 $1.32
gpt-5.6-solopenai/gpt-5.6-sol openai chat $4.40 $22.00
gpt-5.6-terraopenai/gpt-5.6-terra openai chat $2.20 $13.20
gpt-5-miniopenai/gpt-5-mini openai chat $0.25 $2.00
gpt-5-mini-2025-08-07openai/gpt-5-mini-2025-08-07 openai chat $0.25 $2.00
gpt-5-nanoopenai/gpt-5-nano openai chat $0.05 $0.40
gpt-5-nano-2025-08-07openai/gpt-5-nano-2025-08-07 openai chat $0.05 $0.40
gpt-6.1-solopenai/gpt-6.1-sol openai chat $2.00 $10.00
gpt-6-astraopenai/gpt-6-astra openai chat $10.00 $50.00
gpt-6-lunaopenai/gpt-6-luna openai chat $0.10 $0.50
gpt-6-solopenai/gpt-6-sol openai chat $2.00 $10.00
o3openai/o3 openai chat $2.00 $8.00
o3-2025-04-16openai/o3-2025-04-16 openai chat $2.00 $8.00
o3-miniopenai/o3-mini openai chat $1.10 $4.40
o3-mini-2025-01-31openai/o3-mini-2025-01-31 openai chat $1.10 $4.40
o4-miniopenai/o4-mini openai chat $1.10 $4.40
o4-mini-2025-04-16openai/o4-mini-2025-04-16 openai chat $1.10 $4.40
text-embedding-3-largeopenai/text-embedding-3-large openai embedding $0.13 —
text-embedding-3-smallopenai/text-embedding-3-small openai embedding $0.02 —
text-embedding-ada-002openai/text-embedding-ada-002 openai embedding $0.10 —
tts-1openai/tts-1 openai speech $15.00 /1M chars —
tts-1-1106openai/tts-1-1106 openai speech $15.00 /1M chars —
tts-1-hdopenai/tts-1-hd openai speech $30.00 /1M chars —
tts-1-hd-1106openai/tts-1-hd-1106 openai speech $30.00 /1M chars —
whisper-1openai/whisper-1 openai transcription $0.006 /min —

Pricing

Pay for what each call costs

Every call is metered at the catalog price above and the exact cost comes back in the response.

No subscription

Models has no plan of its own. Thundera apps call it with their own keys; the AI allowance in each app plan is enforced here as a per-user monthly budget.

Metered per call

Platform keys are charged at the per-token, per-minute or per-character prices in the catalog. Nothing is marked up silently, and unpriced models are never served.

Hard budgets

Every key has a USD cap enforced before the call. Your plan, cards and invoices are managed centrally in your Thundera account.

See your plans Model prices

Quickstart

Point your SDK at the gateway

The request is plain OpenAI. The only addition is cost in the answer.

curl
curl https://models.thundera.eu/v1/chat/completions \
  -H "Authorization: Bearer $MODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"fast","messages":[{"role":"user","content":"Hello"}]}'
Node.js (openai)
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://models.thundera.eu/v1",
  apiKey: process.env.MODELS_API_KEY,
});
const res = await client.chat.completions.create({
  model: "fast",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(res.choices[0].message.content, res.cost.total_usd);

For Thundera apps

In-cluster, no limits, one key per app

Every platform-funded LLM, embedding and audio call goes through the gateway with the app's own key.

  • No CloudFront limits: all headers pass, long requests are not rewritten
  • Attribute calls to your users with X-Thundera-User and their plan budget
  • Key in SSM and as a Kubernetes Secret, revocable on the next call

App onboarding

bash
# env injected from Secret <app>-models
MODELS_BASE_URL=http://models.thundera.svc.cluster.local/v1
MODELS_API_KEY=<from SSM /thundera/<app>/models-api-key>

# NetworkPolicy egress
to: podSelector app=models  port: 8080

Security

Keys you can revoke, logs you can trust

Database-authoritative keys

RS256 JWTs, verifiable offline via JWKS. Regenerating or revoking a key takes effect on the very next call.

Encrypted payloads

Bodies are stored only when a key opts in: AES-256-GCM, at most 64 KB each, 30-day retention. Off for sensitive apps.

Priced or not served

A model without a price is never exposed, so nothing can run outside the budget accounting.

High availability

Three replicas behind a disruption budget and autoscaler; shared ledger in PostgreSQL; graceful drain on deploy.

Build against the gateway

Full reference with quickstarts for curl, Node, Python, Go and Rust.

Open the docs System (admins)