Inference Gateway · Six Modalities · 50+ Models

PHAROS

The inference layer for AI agents that don't sleep.

api.mypharos.ai · liveOpenAI compat30% over upstream$5 minimum top-up

One parameter · three lanes

Tune cost against quality, per request.

No other gateway lets you pivot model class on a single key. Set quality per request. Get the exact tradeoff for that exact call.

Cheapest

$0.003/ image

Open-weights backends. Draft speed, bulk jobs, high-volume pipelines where cost dominates quality.

Stable Diffusion XL · Flux Schnell

~1.4s p50

quality: "cheapest"

Balanced

$0.04/ image

The sensible default. Quality you can ship, cost you can justify. Covers most production traffic.

Flux Pro · Ideogram v2

~3.2s p50

quality: "balanced"

Best

$0.06/ image

Frontier models for customer-facing output. Hero shots, covers, anything that represents your brand.

GPT-Image-1 · Flux Ultra

~5.8s p50

quality: "best"

Routes through 50+ models.

One key. Any modality.

Chat · EmbeddingsSpeech · TranscriptionImage · Video

Six modalities · one base URL

Everything you call OpenAI for. Plus everything you don't.

All routes are OpenAI-compatible. Drop-in replacement, same SDK, same response shapes. Video is async (custom 202 + poll) because nobody returns a 30-second clip in under 30 seconds.

Scroll · six modalities01 → 06

01modality

Chat.

Streaming completions across frontier and open-weight models.

POST /v1/chat/completions

$0.003/ 1K input · streaming

gpt-5.1 · claude-4.6 · llama-4-405b

How do I light a lighthouse?

Start with a beacon. Then route requests by quality.

02modality

Embeddings.

1536-dim vectors for semantic search and retrieval, batched.

POST /v1/embeddings

$0.00013/ 1K tokens

text-embedding-3-large · voyage-3

03modality

Speech.

Text-to-speech, multi-voice, multi-language, low latency.

POST /v1/audio/speech

$0.015/ 1K chars

elevenlabs-v3 · openai-tts-hd

04modality

Transcription.

Speech-to-text with timestamps. Multipart upload, any audio.

POST /v1/audio/transcriptions

$0.006/ minute

whisper-large-v3 · gpt-4o-transcribe

recording · 00:14

[00:00.4] We need a single inference layer for the agent.

[00:06.1] One that doesn't sleep when traffic spikes at 3am.

[00:11.8] Pharos handles all six modalities

05modality

Image.

Generation across SDXL, Flux, GPT-Image-1. quality: param routes the call.

POST /v1/images/generations

$0.003 → $0.06/ image · quality param

flux-pro · gpt-image-1 · sdxl

06modality

Video.

Long-running generation. Returns 202 + poll_url. No timeouts.

POST /v1/videos/generations

$0.40 → $1.20/ second · async

veo-3 · runway-gen4 · kling-2

frame · 048 / 2404.0s @ 60fps

poll_url · 202 accepted~19s remaining

Drop-in replacement

Two lines change. Everything else stays.

Point your OpenAI client at api.mypharos.ai/v1. Swap the key. Done.

api.mypharos.ai/v1

# Install the SDK (placeholder — SDK publish pending)
npm install pharos-sdk

# Or use the OpenAI SDK directly — it's fully compatible
npm install openai

# Set your key
export PHAROS_API_KEY="pharos_sk_live_…"

# Verify the endpoint is reachable
curl https://api.mypharos.ai/v1/models \
  -H "Authorization: Bearer $PHAROS_API_KEY"

The quality parameter is the only Pharos-specific extension. Everything else is verbatim OpenAI.

Routes through

GPT-5.1Claude 4.7Llama 4 405BMistral Large 3Gemini 2.5 ProGrok 3Command R+DeepSeek R1DeepSeek V3Qwen 2.5GPT-4oClaude 4.6Yi Largetext-embedding-3Voyage 3Cohere Embed v4ElevenLabs v3OpenAI TTS HDCartesia SonicPlayHT 2Whisper Large v3GPT-4o TranscribeDeepgram Nova 3AssemblyAI UniversalFlux ProFlux SchnellFlux UltraGPT-Image-1Stable Diffusion XLSD 3.5Ideogram v2Recraft v3Veo 3Runway Gen-4Kling 2Pika 2Minimax VideoLuma Ray 2ReplicateTogether AIFalGroqAnthropicCerebrasHyperbolicLeptonModalOpenAIMistralDeepInfra

Pricing · usage-based

No seat fees. No tier traps. Flat markup.

You pay 30% over what we pay upstream. Routing, retries, one consolidated invoice.

Free trial

$1.00

Credit on signup. No card. Every modality unlocked.

Pay as you go

+30%

Flat markup over upstream cost. No subscriptions.

Top up

Minimum reload when you scale past the trial.