● 27 frontier models · one agentic endpoint

One AI API for every frontier model

Q: What is Speka?

Speka is a unified AI API gateway that provides one OpenAI-compatible endpoint and key for frontier models including DeepSeek V4 Flash, Llama 3.3, Mistral, Kimi K2.6 and FLUX. Developers change the base_url in their existing SDK — no other code changes required.

Q: Is it really OpenAI-compatible?

Yes. We implement the OpenAI Chat Completions, Embeddings and Images endpoints. Point any OpenAI SDK or client at our base URL, set your Speka key, and use any model id from the catalog — no other code changes.

Q: Where do the models run?

Models run on our managed inference infrastructure across multiple capacity providers. We route each request to healthy capacity and fail over automatically, so you get one stable endpoint instead of managing several vendors.

Q: How does billing work?

Every plan includes a monthly usage allowance measured in real model spend. You're billed per token at transparent published rates. Exceed your allowance and you simply pay standard per-token rates — no overage penalties.

Q: Can I set spending and rate limits?

Yes. Set hard spend caps and per-key rate limits from your dashboard, and get alerts as you approach them. Keys can be scoped per project and revoked instantly.

Q: Do you store my prompts?

We transmit prompt content to model providers solely to generate responses. We store request metadata (model, token counts, latency) for analytics and billing, but we do not use your prompts to train models.

Q: What happens on the free plan?

You get $1 of model usage every month, 1 API key, 10 requests per minute and access to all open models — enough to prototype and test. Upgrade only when you need more.

One OpenAI-compatible API key for DeepSeek V4 Flash, Llama 3.3, Mistral, FLUX and more — with native tool-calling and streaming for agents. Usage-based pricing, instant keys, built-in analytics. Zero infrastructure.

Get your API key — free Read the docs

Free to start — $1 of monthly usage on every account. No credit card required.

chat.completions

curl https://speka.me/v1/chat/completions \
  -H "Authorization: Bearer sk-speka-live-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta/llama-3.1-8b-instruct",
    "messages": [{"role":"user","content":"Explain quantum entanglement simply."}]
  }'

Already using the OpenAI SDK? Just change the base_url and key.

What is Speka?

Speka is a unified AI API gateway that gives developers one OpenAI-compatible API key and endpoint (https://speka.me/v1) to call 27 frontier models — including DeepSeek V4 Flash, Llama 3.3 70B, Mistral Nemotron, Nemotron 3 Ultra, GPT-OSS 120B, and FLUX — without creating separate vendor accounts or managing multiple SDKs. The API supports native tool-calling, streaming, and JSON mode for agentic applications. Pricing is usage-based (per token), with a free tier requiring no credit card.

Frontier models from the labs you already trust

DSDeepSeekNVNVIDIAMMistral AIQQwenGGoogleMMetaΦMicrosoftBFBlack Forest Labs

27+

Models available

99.9%

Uptime target

<1s

Time to first token*

100%

OpenAI-compatible

Why Speka

Built for agents, ready for production

Skip the infrastructure, the rate-limit headaches and the per-vendor SDKs. One gateway, one bill, full observability — so you can ship agents, not plumbing.

Every model, one key

Reasoning, chat, code, vision, embeddings and image generation — all behind a single endpoint and key.

Agent-native API

Native tool-calling, streaming and JSON mode. Point the OpenAI SDK at our base URL — no rewrites, no lock-in.

Predictable rate limits

Per-plan RPM and concurrency, enforced fairly with clear headers so you never get surprised.

Usage-based pricing

Pay per token. Monthly included usage on every plan plus pay-as-you-go overage. No lock-in.

Built-in analytics

Track spend, tokens and latency by key, model and day — right in your dashboard.

Secure by design

Keys are hashed at rest, scoped per project, and revocable instantly. Row-level isolated data.

Model catalog

Popular models, ready to call

Frontier open models, production-ready and priced transparently.

Browse all 27

DeepSeek V4 Flash

DeepSeek

reasoning

Fast open reasoning model with transparent chain-of-thought. Strong at math, logic and multi-step problem solving, with low latency suitable for production and automation.

deepseek-ai/deepseek-v4-flash

Nemotron 3 Super 120B

NVIDIA

reasoning

NVIDIA's mixture-of-experts reasoning model — 120B total parameters with 12B active, so frontier-class reasoning arrives at small-model latency.

nvidia/nemotron-3-super-120b-a12b

Nemotron 3 Ultra 550B

NVIDIA

chat

NVIDIA's flagship open model: 550B parameters with 55B active per token. Frontier quality for demanding generation, analysis and agentic work.

nvidia/nemotron-3-ultra-550b-a55b

Llama 3.1 8B Instruct

Mistral Nemotron

Mistral AI

chat

Mistral and NVIDIA's joint instruct model: excellent function calling, 80+ languages and consistently low latency for production chat.

mistralai/mistral-nemotron

GPT-OSS 20B

OpenAI

code

The small GPT-OSS tier — quick code completion, refactors and shell/tool calls at a fraction of the price.

How it works

Live in three steps

Create an account

Point your SDK

Set the base URL to our gateway and use any model id. Works with the OpenAI SDK, LangChain and Vercel AI SDK.

Ship & scale

Monitor usage, set limits and upgrade as you grow. We handle routing, failover and infrastructure.

Trust & security

Secure by default

Your keys and data are protected at every layer — so you can build with confidence.

Hashed API keys

Keys are hashed at rest and shown only once. Scope them per project and revoke instantly.

Encrypted in transit

All traffic is TLS-encrypted end to end between your app, our gateway and model providers.

Row-level isolation

Your usage, keys and metadata are isolated per account at the database level.

Prompts never trained on

We pass prompts through to generate responses — never to train models. Delete your data anytime.

From the field

Teams ship faster on Speka

“Swapped our OpenAI base URL and had DeepSeek-R1 in prod the same afternoon. The tool-calling just worked.”

A. Nakamura

Staff Engineer

“One bill for every model we test means we stopped juggling six vendor dashboards. Spend is finally legible.”

M. Okafor

Founder, AI startup

“Failover across capacity has kept our agents up through provider outages. That reliability earned our trust.”

J. Petrov

Platform Lead

Pricing

Start free. Scale when you do.

Every plan includes monthly usage. Only pay more when you outgrow it.

Free

$0.00 /mo

Kick the tires. No card required.

Start free

$1 of model usage / month
10 requests / minute
1 API key
Access to all open models

Starter

$19.00 /mo

For indie hackers and side projects.

Get Starter

$25 of model usage included
60 requests / minute
5 API keys
All models incl. vision & image

Pro

$99.00 /mo

For production apps and growing teams.

Get Pro

$150 of model usage included
300 requests / minute
25 API keys
Priority routing & lower latency

Scale

$399.00 /mo

For high-volume, latency-sensitive workloads.

Get Scale

$750 of model usage included
1,200 requests / minute
200 API keys
Highest-priority routing

See the full breakdown on the pricing page.

How Speka compares

Speka vs OpenRouter vs Together AI

Speka is a lean, OpenAI-compatible gateway: 16 frontier models from 7 labs, transparent pay-per-token pricing with no overage penalties, and a genuinely free tier needing no card. OpenRouter and Together AI offer far broader catalogs, but Speka trades breadth for a curated, predictable, drop-in-simple developer experience.

Feature	Speka	OpenRouter	Together AI
OpenAI-compatible endpoint	Yes — drop-in	Yes — documented drop-in	Yes (some OpenAI endpoints unsupported)
Model breadth	16 curated models, 7 labs	300+ models, many providers	200+ open/partner models
Native tool / function calling	Yes	Yes (model-dependent)	Yes (model-dependent)
Built-in per-key analytics	Yes	Yes (usage dashboard)	Partial (usage/spend per key)
Image generation	Yes (FLUX.1 dev/schnell)	Yes	Yes (FLUX, video, audio)
Free tier, no credit card	Yes — $1 usage, no card	Yes — free models, rate-limited	Varies — trial credit
Pay-per-token, no overage penalty	Yes	Yes — pay-per-token	Yes — pay-per-token
Embeddings endpoint	Yes (NV-EmbedQA, NV-Embed)	Varies	Yes

As of June 2026 — figures change frequently; verify current models, pricing, and limits on each vendor's own site.

In plain terms

How developers use Speka

What is an agentic AI API?

An agentic AI API is an endpoint built for multi-step, tool-calling workflows — not just single-turn chat completions. It supports structured function calls, parallel tool use, and streaming responses so an AI agent can reason, act, and iterate without blocking. Speka implements this natively across every model in the catalog without per-model SDK customization.

Which frontier AI models can I access through Speka?

Speka provides API access to frontier open models across reasoning, chat, code, vision, embeddings, and image generation — including DeepSeek V4 Flash, NVIDIA Nemotron Super 49B, Meta Llama 3.3 70B, Mistral Large 3, Moonshot Kimi K2.6, OpenAI GPT-OSS 120B, and Black Forest Labs FLUX. All are called through a single endpoint using standard OpenAI message formatting.

Can I use Speka with the OpenAI SDK?

Yes. Change base_url to https://speka.me/v1 and replace your OpenAI key with your Speka key. No other code changes are required — Speka works with the OpenAI Python and Node.js SDKs, LangChain, the Vercel AI SDK, and any library that accepts a configurable base URL.

Is Speka's AI API production-ready?

Speka targets 99.9% uptime, publishes a live system status page, enforces per-plan rate limits with transparent headers, hashes API keys at rest, scopes them per project, and isolates data at the row level. Predictable concurrency limits and clear overage signals make it suitable for high-throughput production applications and agents.

FAQ

Frequently asked questions

What is Speka?

Speka is a unified AI API gateway that provides one OpenAI-compatible endpoint and key for frontier models including DeepSeek V4 Flash, Llama 3.3, Mistral, Kimi K2.6 and FLUX. Developers change the base_url in their existing SDK — no other code changes required.

Is it really OpenAI-compatible?

Yes. We implement the OpenAI Chat Completions, Embeddings and Images endpoints. Point any OpenAI SDK or client at our base URL, set your Speka key, and use any model id from the catalog — no other code changes.

Where do the models run?

Models run on our managed inference infrastructure across multiple capacity providers. We route each request to healthy capacity and fail over automatically, so you get one stable endpoint instead of managing several vendors.

How does billing work?

Every plan includes a monthly usage allowance measured in real model spend. You're billed per token at transparent published rates. Exceed your allowance and you simply pay standard per-token rates — no overage penalties.

Can I set spending and rate limits?

Yes. Set hard spend caps and per-key rate limits from your dashboard, and get alerts as you approach them. Keys can be scoped per project and revoked instantly.

Do you store my prompts?

We transmit prompt content to model providers solely to generate responses. We store request metadata (model, token counts, latency) for analytics and billing, but we do not use your prompts to train models.

What happens on the free plan?

You get $1 of model usage every month, 1 API key, 10 requests per minute and access to all open models — enough to prototype and test. Upgrade only when you need more.

Get your free AI API key

Join developers shipping faster with Speka. Your first key is free and takes 30 seconds.

Create free account Compare plans