The fastest agentic AI models. One API.

Run DeepSeek, GLM and MiMo UltraSpeed through a single OpenAI-compatible endpoint, with the full upstream cache discount passed through. Cache hits from $0.0077 per million tokens.

Works with Hermes Agent, OpenClaw, OpenCode and other OpenAI-compatible tools.

Get API credit Browse models

client.ts

Drop in

// Before: your existing OpenAI client
const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY })

// After: point at above.dev, keep everything else
const client = new OpenAI({
  apiKey: process.env.ABOVE_API_KEY,
  baseURL: "https://api.above.dev/v1"
})

1000 TOK/S

Speed First

The fastest agentic models

MiMo V2.5 Pro UltraSpeed peaks at 1,000 tokens/second. GLM 5.2 Fast runs at roughly twice standard throughput. Built for agents that cannot wait around.

Browse models

Direct Access

Choose your model

Call any model directly by name: DeepSeek V4 Flash and Pro, GLM 5.2 standard and Fast, MiMo V2.5 Pro UltraSpeed. Buy dollar packs, no subscriptions, full control.

Browse models

Cached tokens, actually cheap

The token class agents actually pay for

Agent sessions re-bill the same prompt on every tool call, so 80 to 95% of your bill is cached input. We pass through DeepSeek's 97% cache discount and Fireworks' 10x GLM discount in full: cache hits from $0.0077/M, where many aggregator providers charge $0.10 to $0.33/M. Everything is upstream cost plus a flat 10%, on every token class.

See pricing

Drop-in

OpenAI-compatible endpoint

Point your existing OpenAI client at https://api.above.dev/v1 and keep everything else. Streaming, tool calls, and structured output all work out of the box.

Read the docs

Live models. Honest pricing.

Direct access to the current above.dev catalog, with separate cache-hit pricing where upstream providers report it.

Where two prices are shown, the model uses off-peak and peak rates. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays only; all other hours, including the whole weekend from 16:00 UTC Friday to 16:00 UTC Sunday, bill at the lower off-peak rate. Every response returns its exact cost in the x-cost-usd header.

What a real agent session costs

A coding-agent session with 50+ tool calls re-sends the same prompt on every call, so most of the input is served from the provider's cache. Example: 1M input tokens (90% cached) plus 100k output on GLM 5.2 costs about $0.78 here, against about $0.81 at zero-markup resellers of the official API. They resell a 5x cache discount at $0.26/M cached; we pass through Fireworks' 10x discount at $0.154/M. A flat 10% on the right upstream beats 0% on the wrong one, and the gap widens the more agentic your traffic is.

DeepSeek V4 FlashCheapest

T1DeepSeek1 million

$ In / 1M: $0.242–$0.484
$ Out / 1M: $0.726–$1.452
Cached input / 1M: $0.0077–$0.0154

DeepSeek V4 Flash Vision (Exp)

T1DeepSeek1 million

$ In / 1M: $0.242–$0.484
$ Out / 1M: $0.726–$1.452
Cached input / 1M: $0.0077–$0.0154

DeepSeek V4 Pro

T2DeepSeek1 million

$ In / 1M: $0.726–$1.452
$ Out / 1M: $2.178–$4.356
Cached input / 1M: $0.0242–$0.0484

GLM 5.2

T2Fireworks1 million

$ In / 1M: $1.54
$ Out / 1M: $4.84
Cached input / 1M: $0.154

GLM 5.2 FastFast router

T3Fireworks1 million

$ In / 1M: $2.31
$ Out / 1M: $7.26
Cached input / 1M: $0.231

Qwen 3.8 Max

T3Fireworks1 million

$ In / 1M: $2.20
$ Out / 1M: $6.60
Cached input / 1M: $0.275

MiMo V2.5 Pro

T2Xiaomi MiMo1 million

$ In / 1M: $0.5077
$ Out / 1M: $1.0154
Cached input / 1M: $0.0042

MiMo V2.5 Pro UltraSpeedUltraSpeed

T3Xiaomi MiMo1 million

$ In / 1M: $1.5231
$ Out / 1M: $3.0462
Cached input / 1M: $0.0127

Model	Provider	Tier	Context	$ In / 1M tokens	$ Cached in / 1M	$ Out / 1M tokens
DeepSeek V4 FlashCheapest	DeepSeek	T1	1 million	$0.242–$0.484	$0.0077–$0.0154	$0.726–$1.452
DeepSeek V4 Flash Vision (Exp)	DeepSeek	T1	1 million	$0.242–$0.484	$0.0077–$0.0154	$0.726–$1.452
DeepSeek V4 Pro	DeepSeek	T2	1 million	$0.726–$1.452	$0.0242–$0.0484	$2.178–$4.356
GLM 5.2	Fireworks	T2	1 million	$1.54	$0.154	$4.84
GLM 5.2 FastFast router	Fireworks	T3	1 million	$2.31	$0.231	$7.26
Qwen 3.8 Max	Fireworks	T3	1 million	$2.20	$0.275	$6.60
MiMo V2.5 Pro	Xiaomi MiMo	T2	1 million	$0.5077	$0.0042	$1.0154
MiMo V2.5 Pro UltraSpeedUltraSpeed	Xiaomi MiMo	T3	1 million	$1.5231	$0.0127	$3.0462

Simple pricing. No subscriptions.

Buy prepaid credit and spend it directly on tokens. No seats, no monthly commitment.

Starter

$10

For testing and small projects

Popular

Builder

$20

For active development

Scale

$50

For production workloads

Credits are metered in micro-dollars at the per-model rates above.

Add credit.
Start calling models.

Checkout creates your account, sends your API key, and opens the dashboard for usage and balance history.

Buy credit Read docs