Pricing

Flat-rate plans with rolling request windows and wallet spillover — plus per-model API rates synced live with our catalog for pay-as-you-go usage.

Flat-rate plans

One price, a deep rolling request window, and no meter spinning in the corner of your eye. Every plan runs on recycled GPUs in regions with access to renewable energy.

Beta pricing, locked in for as long as you stay subscribed.

Creative

$10/mo

For writers, roleplayers, and worldbuilders. The most requests per dollar on Cobble — 1,000 per rolling window, a full stable of fast models tuned for prose and character, and 32K of memory so long stories stay coherent. Your story doesn't stop at a daily cap.

1,000 requests / rolling 5h window

1 concurrency slot

32K context

Models

gemma4/gemma4-12bgemma4/gemma4-26b-a4bgemma4/gemma4-e2bgemma4/gemma4-e4bmistral/ministral-3bmistral/ministral-8bmistral/mistral-nemo-12bopenai/gpt-oss-20bornith/ornith-9bqwen/qwen3.5-9b
Request invite
Most popular

Agent Runner

$20/mo

For assistants that never log off. Flagship-class reasoning (Qwen3.5 122B) with a reserved slot — consistent latency at 3pm and 3am, no queue. Tool calling on every model, 128K context, and 50M embedding tokens a month for agent memory. Built for software that works while you sleep.

750 requests / rolling 5h window

1 concurrency slot — reserved/priority slot

128K context

50M embedding tokens / mo

Everything in Creative

New at this tier

gemma4/gemma4-31bqwen/qwen3.5-122b-a10bqwen/qwen3.6-27b
opencodeHermes AgentOpenClawCursor
Request invite

Coding

$40/mo

$35/seat for teams (3-seat min)

For repo-scale work. DeepSeek V4 Flash and the strongest coding models on the catalog, two parallel slots so agentic tool calls don’t queue, 160K context, and wallet spillover so a long session never hits a wall you didn’t place. Works out of the box with opencode, Hermes Agent, OpenClaw, and Cursor.

600 requests / rolling 5h window

2 concurrency slots

160K context

Everything in Agent Runner

New at this tier

deepseek/deepseek-v4-flashornith/ornith-1-31bornith/ornith-1-35bqwen/qwen3.6-35b-a3b
opencodeHermes AgentOpenClawCursor
Request invite

A La Carte

No subscription — fund a wallet, pay published rates per token. Top-ups never expire.

Top-up presets

$10$25$50$100

Custom amounts $5–$500

How limits work

  • Rolling window — Each plan includes a 5-hour trailing request allowance (Creative: 1,000 · Agent Runner: 750 · Coding: 600) that refills continuously as older requests age out — not a daily reset cliff. Details in the docs.
  • Concurrency slots — Parallel in-flight requests are capped per plan (Creative: 1 · Agent Runner: 1 priority · Coding: 2). Extra agent loops queue until a slot opens. See slot behavior.
  • Wallet spillover — When you opt in, a funded wallet covers requests past your rolling window, above-tier models, long context, and embedding overages — debited at published catalog rates with a monthly cap you set. Spillover & billing.

Sustainability, included

Recycled GPUs

Built from reclaimed GPUs and server hardware — extending useful life instead of manufacturing new silicon.

Renewable-ready

Deployed in regions with access to renewable energy.

No evaporative cooling

Designed to avoid evaporative water cooling.

Per-model API rates

Pay-as-you-go token, page, and endpoint pricing — synced live with our catalog.

Generative AI

ModelProviderInput / 1M tokOutput / 1M tokCached / 1M tokPages / 1KEndpoint
Qwen3.5 122B A10BFLAGSHIP
Alibaba Cloud$0.23/1M$1.87/1M$0.4675/1M—/v1/chat/completions
Qwen3.6 27B
Alibaba Cloud$0.25/1M$1.8/1M$0.45/1M—/v1/chat/completions
Qwen3.6 35B A3B
Alibaba Cloud$0.12/1M$0.86/1M$0.215/1M—/v1/chat/completions
Gemma4 31B
Google DeepMind$0.1/1M$0.31/1M$0.0775/1M—/v1/chat/completions
Qwen3.5 9B
Alibaba Cloud$0.09/1M$0.13/1M$0.0325/1M—/v1/chat/completions
Gemma4 26B A4B
Google DeepMind$0.05/1M$0.27/1M$0.0675/1M—/v1/chat/completions
Gemma4 12B
Google DeepMind$0.09/1M$0.27/1M$0.0675/1M—/v1/chat/completions
Gemma4 E4B
Google DeepMind$0.05/1M$0.1/1M$0.025/1M—/v1/chat/completions
Gemma4 E2B
Google DeepMind$0.03/1M$0.06/1M$0.015/1M—/v1/chat/completions
Ministral 8B
Mistral AI$0.135/1M$0.135/1M$0.0338/1M—/v1/chat/completions
Ministral 3B
Mistral AI$0.09/1M$0.09/1M$0.0225/1M—/v1/chat/completions
Mistral Nemo 12B
Mistral AI$0.018/1M$0.027/1M$0.0068/1M—/v1/chat/completions
GPT-OSS 20B
OpenAI$0.026/1M$0.117/1M$0.0293/1M—/v1/chat/completions
DeepSeek V4 FlashDeepSeek
Coming soon
Ornith 9BDeepReinforce AI
Coming soon
Ornith 1 35BDeepReinforce AI
Coming soon
Ornith 1 31BDeepReinforce AI
Coming soon

OCR

ModelProviderInput / 1M tokOutput / 1M tokCached / 1M tokPages / 1KEndpoint
Chandra OCRFEATURED
Cobble Labs———$4/v1/ocr
GLM-OCR
Zhipu AI———$0.08/v1/ocr
DeepSeek OCR2
DeepSeek———$0.08/v1/ocr
Nemotron OCR v2
NVIDIA———$0.08/v1/ocr

Embeddings

ModelProviderInput / 1M tokOutput / 1M tokCached / 1M tokPages / 1KEndpoint
Nomic Embed 1.5FEATURED
Nomic AI$0.01/1M———/v1/embeddings
Granite Embedding 311M
IBM$0.1/1M———/v1/embeddings
Granite Embedding 97M
IBM$0.09/1M———/v1/embeddings
Qwen3 Embedding 0.6B
Alibaba Cloud$0.01/1M———/v1/embeddings
Qwen3 Embedding 8B
Alibaba Cloud$0.01/1M———/v1/embeddings

Wallet top-up

Fund your wallet for spillover and pay-as-you-go usage. Top-ups never expire.

FAQ

How does the rolling request window work?

Each plan includes a trailing 5-hour request allowance that refills continuously as older requests age out — there is no daily reset or monthly credit balance to track.

What happens when I exceed my window?

If you opt into wallet spillover, requests past your rolling window (plus above-tier models, long context, and embedding overages) are debited from your wallet at published catalog rates. Otherwise they queue until the window refills. Wallet top-ups never expire.

How is inference cost calculated?

Flat-rate plans cover usage within your rolling window. Anything billed to your wallet uses the per-model token, page, and endpoint rates listed in the tables above. Failed requests are not charged where metering supports it.

Can I change plans?

Yes. Upgrade or downgrade through billing settings in your account. Changes typically take effect on the next cycle per Stripe rules.

What is the concurrent request limit?

It caps how many API requests can run in parallel per key. Higher tiers allow more throughput for bursty workloads.

Questions? Documentation · Platform