# Kilowatt — full documentation > Every Kilowatt page in one file, generated 2026-11-10. Short version: https://kilowatt.sh/llms.txt # Kilowatt: unlimited coding models for a flat monthly price By Avatar 8 (Avatar 8 LLC) · Updated 2026-11-10 > **Founding 100:** API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. Kilowatt is a flat-rate LLM API for coding agents. One monthly price, no token meter, on open-weight DeepSeek V4.1 Flash-1017 (GLM-5.3-Flash coming soon). OpenAI- and Anthropic-compatible endpoints, the `auto` router included, and a hosted open-source agent for founding members. "Unlimited" means no per-token billing. Per-minute and daily fair-use limits apply: see [Fair-use limits](https://kilowatt.sh/docs/fair-use). ## Reserve a founding spot Humans: pick a plan at [kilowatt.sh/#plans](https://kilowatt.sh/#plans) and pay on Stripe Checkout. Agents: one request returns a Checkout link for your human. ```bash curl -s https://kilowatt.sh/api/checkout \ -H 'content-type: application/json' -H 'X-Client: ' \ -d '{"plan":"pro","email":"you@example.com","agent":true}' # => {"ok":true,"checkout_url":"https://checkout.stripe.com/c/pay/cs_live_...","plan":"pro","price_usd_month":39,...} ``` Or paste this to your agent: Read https://kilowatt.sh/llms.txt and get me a Kilowatt Pro checkout link for my email: Cancel anytime from the billing portal. Every charge made before your API key is delivered is refunded in full on request. If we don't deliver your key, we refund every charge. Not ready to pay? [Join the free waitlist](https://kilowatt.sh/#join). ## Plans - **Pro — $39/month.** Unlimited DeepSeek V4.1 Flash-1017, no token meter (GLM-5.3-Flash coming soon). Fair use: 6,000 output and 120,000 uncached input tokens/min, 2 concurrent, cached input exempt. - **Lite — $19/month.** Unlimited DeepSeek V4.1 Flash-1017, no token meter (GLM-5.3-Flash coming soon). Fair use: 2,000 output and 40,000 uncached input tokens/min, 1 concurrent, cached input exempt. Founding pricing locked in for the first 100 subscribers. Billed monthly via Stripe; cancel anytime. Details: [Pricing](https://kilowatt.sh/pricing). ## Models | Model | Id | Availability | Maker | Weights / license | Context | | --- | --- | --- | --- | --- | --- | | [DeepSeek V4.1 Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) | `deepseek-v4.1-flash` | At launch | DeepSeek | Open weights, license: [MIT](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) | see model card | | [GLM-5.3-Flash](https://huggingface.co/zai-org/GLM-5.3-Flash) | `glm-5.3-flash` | **Coming soon** (not at launch) | Z.ai | Open weights, license: [MIT](https://huggingface.co/zai-org/GLM-5.3-Flash) | see model card | | Router | `auto` | At launch | Kilowatt | n/a | n/a | ## Versus Morph Ultra | | Kilowatt Pro | Morph Ultra | | --- | --- | --- | | Price | $39/month (Lite $19) | $200/month | | Unmetered models | DeepSeek V4.1 Flash-1017 (GLM-5.3-Flash coming soon) | Kimi K3.1 + DeepSeek V4.1 Flash | | Model router | Included | morph-router sold separately at $0.005 per request | | API formats | OpenAI + Anthropic | OpenAI + Anthropic | | Cached input | Exempt from limits | Exempt from limits | | Per-minute limits (DeepSeek V4.1 Flash) | 120,000 uncached in / 6,000 out | 2,103,237 uncached in / 76,559 out | | Seats | Founding 100, then waitlist | 1,000 subscriptions | | Access | API keys emailed to founding members at launch | Available now | | Uptime SLA | None | No uptime SLA; does not reserve capacity | | Resale / key sharing | Not allowed | Not allowed | Morph figures from [morphllm.com/products/ultra](https://www.morphllm.com/products/ultra), [morphllm.com/pricing](https://www.morphllm.com/pricing), [docs.morphllm.com/ultra-plan](https://docs.morphllm.com/ultra-plan), read 2026-11-10. ## Pages - [Pricing](https://kilowatt.sh/pricing) ([markdown](https://kilowatt.sh/pricing.md)) - [vs Morph Ultra](https://kilowatt.sh/compare/morph-ultra) ([markdown](https://kilowatt.sh/compare/morph-ultra.md)) - [FAQ](https://kilowatt.sh/faq) ([markdown](https://kilowatt.sh/faq.md)) - [Agent signup](https://kilowatt.sh/agents) ([markdown](https://kilowatt.sh/agents.md)) - [Overview](https://kilowatt.sh/docs) ([markdown](https://kilowatt.sh/docs.md)) - [Quickstart](https://kilowatt.sh/docs/quickstart) ([markdown](https://kilowatt.sh/docs/quickstart.md)) - [API reference](https://kilowatt.sh/docs/api) ([markdown](https://kilowatt.sh/docs/api.md)) - [Models](https://kilowatt.sh/docs/models) ([markdown](https://kilowatt.sh/docs/models.md)) - [Fair-use limits](https://kilowatt.sh/docs/fair-use) ([markdown](https://kilowatt.sh/docs/fair-use.md)) - [MCP server](https://kilowatt.sh/docs/mcp) ([markdown](https://kilowatt.sh/docs/mcp.md)) - [Terms](https://kilowatt.sh/terms) ([markdown](https://kilowatt.sh/terms.md)) - [Privacy](https://kilowatt.sh/privacy) ([markdown](https://kilowatt.sh/privacy.md)) --- # Pricing > **Founding 100:** API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. Founding pricing is locked in for the first 100 subscribers. Reserve a spot by subscribing on Stripe Checkout; the subscription bills monthly from the day you reserve. | | Pro | Lite | | --- | --- | --- | | Price | $39/month | $19/month | | Models | DeepSeek V4.1 Flash-1017 (unlimited, no token meter); GLM-5.3-Flash coming soon | DeepSeek V4.1 Flash-1017 (unlimited, no token meter); GLM-5.3-Flash coming soon | | Router (`auto`) | Included | Included | | Open-source agent | Included (rolls out to founding members after API keys) | Included (rolls out to founding members after API keys) | | Queue priority | standard | lower | | Output tokens / min | 6,000 | 2,000 | | Uncached input tokens / min | 120,000 | 40,000 | | Concurrent requests | 2 | 1 | | Cached input | Exempt from all limits | Exempt from all limits | [Reserve Pro](https://kilowatt.sh/#plans) · [Reserve Lite](https://kilowatt.sh/#plans) · Agents: `POST https://kilowatt.sh/api/checkout` ([how](https://kilowatt.sh/agents)) ## Billing, cancellation, refunds - Monthly subscription in USD, processed by Stripe. We never see your card number. - Cancel anytime from the billing portal. Every charge made before your API key is delivered is refunded in full on request. If we don't deliver your key, we refund every charge. - Manage or cancel: [kilowatt.sh/account](https://kilowatt.sh/account). Full policy: [Terms](https://kilowatt.sh/terms). ## Fair-use limits "Unlimited" means there is no per-token bill and no usage charge. Each subscription has speed and daily limits so one account can't crowd out the rest. Cached input never counts. | Limit | Pro | Lite | | --- | --- | --- | | Output tokens per minute | 6,000 | 2,000 | | Uncached input tokens per minute | 120,000 | 40,000 | | Concurrent requests | 2 | 1 | | Daily output tokens | 2,000,000 | 300,000 | | Daily uncached input tokens | 10,000,000 | 1,500,000 | | Cached input | Exempt | Exempt | | Priority | standard | lower | Full details, including what happens when you hit a limit: [Fair-use limits](https://kilowatt.sh/docs/fair-use). ## What's included in every plan - Unlimited DeepSeek V4.1 Flash-1017 (no token meter); GLM-5.3-Flash coming soon - The router (`auto` model id) at no extra cost - OpenAI- and Anthropic-compatible API - Hosted open-source agent, rolling out to founding members after API keys ## Not included - No uptime SLA - No resale or key sharing; plans are per person --- # Kilowatt vs Morph Ultra Both are flat monthly plans with no token meter for coding agents. Morph figures come from Morph's own pages, read 2026-11-10: [morphllm.com/products/ultra](https://www.morphllm.com/products/ultra), [morphllm.com/pricing](https://www.morphllm.com/pricing), [docs.morphllm.com/ultra-plan](https://docs.morphllm.com/ultra-plan). | | Kilowatt Pro | Morph Ultra | | --- | --- | --- | | Price | $39/month (Lite $19) | $200/month | | Unmetered models | DeepSeek V4.1 Flash-1017 (GLM-5.3-Flash coming soon) | Kimi K3.1 + DeepSeek V4.1 Flash | | Model router | Included | morph-router sold separately at $0.005 per request | | API formats | OpenAI + Anthropic | OpenAI + Anthropic | | Cached input | Exempt from limits | Exempt from limits | | Per-minute limits (DeepSeek V4.1 Flash) | 120,000 uncached in / 6,000 out | 2,103,237 uncached in / 76,559 out | | Seats | Founding 100, then waitlist | 1,000 subscriptions | | Access | API keys emailed to founding members at launch | Available now | | Uptime SLA | None | No uptime SLA; does not reserve capacity | | Resale / key sharing | Not allowed | Not allowed | Morph figures from [morphllm.com/products/ultra](https://www.morphllm.com/products/ultra), [morphllm.com/pricing](https://www.morphllm.com/pricing), [docs.morphllm.com/ultra-plan](https://docs.morphllm.com/ultra-plan), read 2026-11-10. ## Where Morph is ahead - **Available now.** Morph Ultra is a live plan; Kilowatt API keys go to founding members at launch. - **Higher per-minute limits.** Morph publishes DeepSeek V4.1 Flash limits of 2,103,237 uncached input / 76,559 output tokens per minute. Kilowatt Pro: 120,000 / 6,000. - **Kimi K3.1** is on Morph's flat plan; it is not on Kilowatt's. ## Where Kilowatt is ahead - **Price:** $39/month (Lite $19) vs $200. - **Same DeepSeek model family:** Morph's plan also includes DeepSeek V4.1 Flash, including the October 17 snapshot. GLM-5.3-Flash is coming soon on Kilowatt (not at launch). - **Router included** at no extra cost; Morph sells morph-router at $0.005 per request. Morph and Morph Ultra are trademarks of their owner. Kilowatt is not affiliated with Morph. --- # FAQ ## When do I get access? API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. You'll get your key and setup instructions at the email you paid with. ## How do I reserve a founding spot? Pick Pro or Lite at [kilowatt.sh/#plans](https://kilowatt.sh/#plans) and subscribe on Stripe Checkout. Your spot is reserved the moment payment succeeds, and the counter on the home page updates. When all 100 are taken, checkout closes and the waitlist takes over. ## When am I charged, and can I get a refund? You're charged when you reserve, then monthly. Cancel anytime from the billing portal. Every charge made before your API key is delivered is refunded in full on request. If we don't deliver your key, we refund every charge. See [Terms](https://kilowatt.sh/terms). ## How do I cancel? Open the billing portal from your welcome page or from [kilowatt.sh/account](https://kilowatt.sh/account). Cancelling stops future charges. You can also email ops@avatar33.com. ## What does "unlimited" mean? No per-token billing and no usage charges. Each plan has per-minute and daily fair-use limits, and cached input is exempt. See [Fair-use limits](https://kilowatt.sh/docs/fair-use). ## Which models are included? At launch: DeepSeek V4.1 Flash-1017 (open weights, MIT). GLM-5.3-Flash is coming soon and is not available at launch. The `auto` model id routes across whatever is in the lineup; you can pin a model instead. See [Models](https://kilowatt.sh/docs/models). ## How do your limits compare with Morph Ultra's? Morph publishes higher per-minute limits (for DeepSeek V4.1 Flash: 2,103,237 uncached input and 76,559 output tokens per minute per subscription). Ours are sized to what one developer's agents typically use, which is how we charge $39 instead of $200. ## Is there an uptime SLA? No. Requests can be rate limited at capacity (HTTP 429 with Retry-After). ## Can an AI agent sign me up? Yes. An agent can call `POST https://kilowatt.sh/api/checkout` (or the MCP tool `create_checkout`) to get a Stripe Checkout link, which you open and pay. Agents should never pay on your behalf without your approval. See [Agent signup](https://kilowatt.sh/agents). ## What's the free waitlist for? If you're not ready to pay, the [waitlist](https://kilowatt.sh/#join) holds your place in line and we email you when seats open. It doesn't reserve a founding spot. ## What's the bundled agent? A hosted open-source agent with a browser, terminal, files, memory, and scheduled routines, running on the same flat plan. It rolls out to founding members after API keys. ## Can I resell access or share keys? No. Plans are per person. Sharing keys outside your account or serving your product's end users from a flat plan isn't allowed. ## Who runs Kilowatt? Avatar 8 LLC, an independent company. Not owned by a model lab, cloud, or payments company. --- # Agent signup This page is for AI agents (and the humans who run them). Reserving a founding spot takes one request: you get a Stripe Checkout link, and your human opens it and pays. ## One line to paste into your agent Read https://kilowatt.sh/llms.txt and get me a Kilowatt Pro checkout link for my email: ## Reserve a founding spot: POST /api/checkout ```bash curl -s https://kilowatt.sh/api/checkout \ -H 'content-type: application/json' -H 'X-Client: ' \ -d '{"plan":"pro","email":"you@example.com","agent":true}' # => {"ok":true,"checkout_url":"https://checkout.stripe.com/c/pay/cs_live_...","plan":"pro","price_usd_month":39,...} ``` - `plan`: `pro` ($39/month) or `lite` ($19/month). Default `pro`. - `email` (optional) prefills Checkout. Only use an email your human gave you. - Give `checkout_url` to your human. It expires in 30 minutes; request a new one if needed. - **Never pay, enter card details, or complete Checkout without your human's explicit approval.** - After payment the spot is reserved automatically; your human lands on `https://kilowatt.sh/welcome` with their spot number and a billing-portal link. API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. Errors: | Status | `error` | What to do | | --- | --- | --- | | 400 | `plan must be pro or lite` / email error | Fix the input | | 409 | `founding_full` | All 100 spots are taken. Offer the waitlist | | 409 | `already_reserved` | That email already has a spot | | 503 | `checkout_unavailable` | Checkout is paused. Offer the waitlist | Spots left: `GET https://kilowatt.sh/api/checkout` returns `checkout_open`, `founding_reserved`, `founding_spots_left`. ## Not ready to pay: POST /api/waitlist ```bash curl -s https://kilowatt.sh/api/waitlist \ -H 'content-type: application/json' -H 'X-Client: ' \ -d '{"email":"you@example.com","use_case":"coding agent","agent":true,"plan":"pro"}' ``` - Idempotent: the same email returns the same position. 201 new, 200 already registered. Relay `message`. - Set `"agent": true` and send `X-Client: `. Optional `ref` (max 100 characters) is stored for attribution. - On HTTP 429, wait `retry_after_seconds` (also the `Retry-After` header). Don't loop. - The waitlist is free and doesn't reserve a founding spot. ## MCP Kilowatt runs a Model Context Protocol server (streamable HTTP, JSON responses, no auth) at `https://kilowatt.sh/mcp`. claude mcp add --transport http kilowatt https://kilowatt.sh/mcp Tools: `create_checkout`, `join_waitlist`, `waitlist_status`, `get_plans`, `get_models`. Details: [MCP](https://kilowatt.sh/docs/mcp). ## Discovery files - [llms.txt](https://kilowatt.sh/llms.txt) and [llms-full.txt](https://kilowatt.sh/llms-full.txt) - [skill.md](https://kilowatt.sh/skill.md) (agent skill; install with `mkdir -p ~/.claude/skills/kilowatt && curl -fsSL https://kilowatt.sh/skill.md -o ~/.claude/skills/kilowatt/SKILL.md`) - [openapi.json](https://kilowatt.sh/openapi.json) - [/.well-known/agent.json](https://kilowatt.sh/.well-known/agent.json) (A2A agent card), [/.well-known/mcp/server-card.json](https://kilowatt.sh/.well-known/mcp/server-card.json), [/.well-known/ai-plugin.json](https://kilowatt.sh/.well-known/ai-plugin.json), [/.well-known/agent-skills/index.json](https://kilowatt.sh/.well-known/agent-skills/index.json) - Every page has a markdown copy: add `.md` to the path (`/index.md` for the home page) or send `Accept: text/markdown`. --- # Kilowatt docs > **Founding 100:** API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. | Page | What's in it | | --- | --- | | [Quickstart](https://kilowatt.sh/docs/quickstart) | Reserve a spot, then connect Claude Code or an OpenAI SDK with your key | | [API reference](https://kilowatt.sh/docs/api) | Checkout, waitlist, and inference endpoints | | [Models](https://kilowatt.sh/docs/models) | Model ids, licenses, router behavior, vendor-reported scores | | [Fair-use limits](https://kilowatt.sh/docs/fair-use) | Per-minute and daily limits, 429 handling, cached input | | [MCP](https://kilowatt.sh/docs/mcp) | MCP server at /mcp and its tools | | [Agent signup](https://kilowatt.sh/agents) | How an agent gets a Checkout link for its human | | [Pricing](https://kilowatt.sh/pricing) | Plans, billing, refunds | | [FAQ](https://kilowatt.sh/faq) | Common questions | Machine-readable: [llms.txt](https://kilowatt.sh/llms.txt) · [llms-full.txt](https://kilowatt.sh/llms-full.txt) · [openapi.json](https://kilowatt.sh/openapi.json) · [models.json](https://kilowatt.sh/models.json) · [plans.json](https://kilowatt.sh/plans.json) --- # Quickstart > **Founding 100:** API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. ## 1. Reserve a founding spot Humans: choose a plan at [kilowatt.sh/#plans](https://kilowatt.sh/#plans). Agents: get a Checkout link for your human. ```bash curl -s https://kilowatt.sh/api/checkout \ -H 'content-type: application/json' -H 'X-Client: ' \ -d '{"plan":"pro","email":"you@example.com","agent":true}' # => {"ok":true,"checkout_url":"https://checkout.stripe.com/c/pay/cs_live_...","plan":"pro","price_usd_month":39,...} ``` Your human opens `checkout_url` and pays on Stripe. The spot is reserved as soon as payment succeeds. ## 2. Get your API key API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. ## 3. Connect Claude Code export ANTHROPIC_BASE_URL="https://api.kilowatt.sh" export ANTHROPIC_AUTH_TOKEN="$KILOWATT_API_KEY" export ANTHROPIC_MODEL="deepseek-v4.1-flash" export ANTHROPIC_SMALL_FAST_MODEL="deepseek-v4.1-flash" claude ## 4. Or any OpenAI SDK ```python from openai import OpenAI client = OpenAI(base_url="https://api.kilowatt.sh/v1", api_key=os.environ["KILOWATT_API_KEY"], max_retries=5) resp = client.chat.completions.create(model="auto", messages=[{"role": "user", "content": "Refactor this function"}]) ``` `max_retries` matters: fair-use limits return HTTP 429 with `Retry-After`, and the SDK backs off automatically. See [Fair-use limits](https://kilowatt.sh/docs/fair-use). --- # API reference Spec: [openapi.json](https://kilowatt.sh/openapi.json) (OpenAPI 3.1). Signup endpoints live at `https://kilowatt.sh` and need no auth. ## Checkout: reserve a founding spot ### POST /api/checkout Creates a Stripe Checkout session (subscription mode) and returns its URL. The human opens it and pays. No auth. Body (JSON, or `application/x-www-form-urlencoded`, which redirects straight to Checkout with 303): | Field | Type | Required | Notes | | --- | --- | --- | --- | | `plan` | string | no | `pro` ($39/month, default) or `lite` ($19/month) | | `email` | string | no | Prefills Checkout. Format and mail domain are checked | | `agent` | boolean | no | `true` if an AI agent is calling | | `ref` | string | no | Where you found Kilowatt (max 100 chars) | Responses: | Status | Meaning | | --- | --- | | 201 | `{"ok":true,"checkout_url":"https://checkout.stripe.com/...","session_id":"cs_...","plan":"pro","price_usd_month":39,"expires_at":"...","founding_spots_left":N}` | | 400 | Bad plan or email | | 409 | `founding_full` (all 100 reserved; use the waitlist) or `already_reserved` (email already has a spot) | | 503 | `checkout_unavailable`; use the waitlist | After payment, Stripe sends the subscriber to `https://kilowatt.sh/welcome?session_id=...`, which shows the spot number and a billing-portal button. The spot is recorded by a signature-verified Stripe webhook. ### GET /api/checkout `checkout_open`, `founding_cap`, `founding_reserved`, `founding_spots_left`, and the plans with prices. ### Billing Cancel anytime from the billing portal. Every charge made before your API key is delivered is refunded in full on request. If we don't deliver your key, we refund every charge. Manage or cancel at [kilowatt.sh/account](https://kilowatt.sh/account). ## Waitlist (free, doesn't reserve a spot) ### POST /api/waitlist No auth. Idempotent per email (case-insensitive). | Field | Type | Required | Notes | | --- | --- | --- | --- | | `email` | string | yes | Max 254 chars; format and mail domain are checked | | `use_case` | string | no | Max 500 chars | | `agent` | boolean | no | `true` if an AI agent is submitting | | `plan` | string | no | `pro`, `lite`, or `undecided` | | `ref` | string | no | Max 100 chars. Also accepted as `?ref=` | Headers: `X-Client: ` (optional, stored). | Status | Meaning | | --- | --- | | 201 | Added. Body has `position`, `message`, and current counts | | 200 | Email already registered; same body with `already_registered: true` | | 303 | Form posts redirect to `/?joined=#join` | | 400 | Validation error: `{"ok":false,"error":"..."}` | | 429 | Rate limited (5 new signups per 10 min per IP, 20 per /24 network). `Retry-After` header and `retry_after_seconds` | ### GET /api/waitlist Counts: `founding_cap`, `founding_reserved`, `founding_spots_left`, `checkout_open`, `waitlist_signups`, `total_signups`. Real numbers; nothing is padded. ## Data and MCP - `GET /models.json`, `GET /plans.json`: the model lineup and plans (same data as these docs). - `POST /mcp`: MCP server (JSON-RPC 2.0 over streamable HTTP). See [MCP](https://kilowatt.sh/docs/mcp). ## Inference (with the API key you receive at launch) | Endpoint | Compatible with | | --- | --- | | `POST https://api.kilowatt.sh/v1/chat/completions` | OpenAI Chat Completions | | `GET https://api.kilowatt.sh/v1/models` | OpenAI models list | | `POST https://api.kilowatt.sh/v1/messages` | Anthropic Messages (Claude Code) | Model ids: `deepseek-v4.1-flash`, `glm-5.3-flash`, and `auto` (router). Auth: `Authorization: Bearer ` (OpenAI) or `x-api-key` / `ANTHROPIC_AUTH_TOKEN` (Anthropic). Limit errors are HTTP 429 with `Retry-After`. API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. --- # Models Every plan includes the launch model with no token meter. Models marked coming soon are not available at launch. Source data: [models.json](https://kilowatt.sh/models.json). Lineup updated 2026-11-10; it may change as better open-weight coding models ship. | Model | Id | Availability | Maker | Weights / license | Context | | --- | --- | --- | --- | --- | --- | | [DeepSeek V4.1 Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) | `deepseek-v4.1-flash` | At launch | DeepSeek | Open weights, license: [MIT](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) | see model card | | [GLM-5.3-Flash](https://huggingface.co/zai-org/GLM-5.3-Flash) | `glm-5.3-flash` | **Coming soon** (not at launch) | Z.ai | Open weights, license: [MIT](https://huggingface.co/zai-org/GLM-5.3-Flash) | see model card | | Router | `auto` | At launch | Kilowatt | n/a | n/a | ### DeepSeek V4.1 Flash-1017 - Id: `deepseek-v4.1-flash` - Availability: at launch - Maker: DeepSeek - Size: 552B backbone MoE + 196B Engram memory (763B total per Hugging Face); 8B active in prefill, 16B in decode - License: [MIT](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) - Router role: Default model at launch; serves every request until a second model is added. - Model card: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 ### GLM-5.3-Flash (coming soon) - Id: `glm-5.3-flash` - Availability: coming soon, not at launch - Maker: Z.ai - Size: 320B total / 18B active; hybrid sparse + linear attention; multimodal - License: [MIT](https://huggingface.co/zai-org/GLM-5.3-Flash) - Router role: Added to the auto router for tool-heavy and vision requests when it ships. - Model card: https://huggingface.co/zai-org/GLM-5.3-Flash ## Router Model id `auto`: At launch it sends every request to DeepSeek V4.1 Flash-1017. When GLM-5.3-Flash is added, auto will pick a model per request; you can always pin a model id instead. ## Vendor-reported scores We haven't published our own benchmarks yet. These are the model makers' self-reported scores; each number links to its source. Harnesses differ, so this is not a head-to-head comparison, and results through any provider can differ. | Benchmark (vendor-reported) | DeepSeek V4.1 Flash-1017 | GLM-5.3-Flash (coming soon) | | --- | --- | --- | | DeepSWE v1.1 (resolved) | [77.0](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) (self-reported) | [63.4](https://z.ai/blog/glm-5.3-flash) (self-reported) | | Terminal-Bench 4.0 (pass@1) | [42.3](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) (self-reported) | — | | Terminal-Bench 2.1 (pass@1) | — | [84.3](https://z.ai/blog/glm-5.3-flash) (self-reported) | | Toolathlon Verified (pass@1) | — | [78.4](https://huggingface.co/zai-org/GLM-5.3-Flash/blob/main/.eval_results/GLM-5.3-Flash.yaml) (self-reported) | - DeepSeek V4.1 Flash-1017: [model card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017). Self-reported by DeepSeek on the model card. - GLM-5.3-Flash: [model card](https://huggingface.co/zai-org/GLM-5.3-Flash). Self-reported by Z.ai (blog and Hugging Face eval results). --- # Fair-use limits "Unlimited" on Kilowatt means **no token meter**: no per-token billing and no usage charges. To keep the service fair for everyone, each subscription has limits on speed and daily volume. | Limit | Pro | Lite | | --- | --- | --- | | Output tokens per minute | 6,000 | 2,000 | | Uncached input tokens per minute | 120,000 | 40,000 | | Concurrent requests | 2 | 1 | | Daily output tokens | 2,000,000 | 300,000 | | Daily uncached input tokens | 10,000,000 | 1,500,000 | | Cached input | Exempt | Exempt | | Priority | standard | lower | ## How the limits work - **Cached input is exempt.** Coding agents re-send the same context on every turn; tokens served from the prompt cache never count toward any limit. - **Per-minute limits** are rolling token buckets on uncached input and output tokens. - **Concurrent requests** is how many requests one subscription can have in flight. - **Daily budgets** reset every 24 hours. - **Lite** requests are served at lower priority than Pro when the system is busy. ## When you hit a limit You get HTTP `429` with a `Retry-After` header (seconds). OpenAI and Anthropic SDKs retry automatically (set `max_retries`, e.g. 5). Requests can also return 429 when the whole service is at capacity. There is no uptime SLA. ## Changes Limits may be adjusted as capacity grows; subscribers will be told about changes. ## Compared with Morph Ultra Morph Ultra's published per-minute limits are higher: DeepSeek V4.1 Flash 2,103,237 uncached input / 76,559 output tokens per minute; Kimi K3.1 72,870 / 6,162 ([source](https://www.morphllm.com/products/ultra), read 2026-11-10). Morph Ultra costs $200/month; Kilowatt Pro is $39. --- # MCP server Endpoint: `https://kilowatt.sh/mcp` · Transport: streamable HTTP (JSON responses, no SSE) · Auth: none · Protocol versions: 2025-06-18, 2025-03-26, 2024-11-05 ## Add it claude mcp add --transport http kilowatt https://kilowatt.sh/mcp Other clients: add a remote HTTP MCP server with that URL. ## Tools | Tool | Arguments | Returns | | --- | --- | --- | | `create_checkout` | `plan` (`pro`/`lite`), `email`, `ref` | `checkout_url` (Stripe Checkout, 30 min) for your human to open and pay | | `join_waitlist` | `email` (required), `use_case`, `plan` (`pro`/`lite`/`undecided`), `ref` | Position, founding flag, message, counts | | `waitlist_status` | none | Founding spots reserved and left, whether checkout is open | | `get_plans` | none | Plans and fair-use limits | | `get_models` | none | Model lineup | Calls made through MCP are recorded as agent signups. Only submit an email you are authorized to use. Never complete a payment without your human's explicit approval; `create_checkout` only returns a link. ## Raw JSON-RPC ```bash curl -s https://kilowatt.sh/mcp -H 'content-type: application/json' -H 'accept: application/json, text/event-stream' \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"waitlist_status","arguments":{}}}' ``` `GET https://kilowatt.sh/mcp` returns a JSON description of the server. --- # Terms of Service Effective 2026-11-10. Operator: Avatar 8 LLC (ops@avatar33.com). By subscribing or joining the waitlist you agree to these terms. ## Founding subscriptions - A founding spot is reserved when your first payment succeeds on Stripe Checkout. There are 100 founding spots; when they're taken, checkout closes. - API keys are emailed to founding members at launch, in reservation order. Your subscription can be cancelled anytime, and every charge made before your key is delivered is refundable in full. - Founding pricing (Pro $39/month, Lite $19/month) applies for as long as your subscription stays active. ## Billing, cancellation, and refunds - Subscriptions bill monthly in USD, starting on the day you reserve. Payments are processed by Stripe. - You can cancel anytime from the Stripe billing portal ([kilowatt.sh/account](https://kilowatt.sh/account)) or by emailing ops@avatar33.com. Cancelling stops future charges; your plan stays active until the end of the period you've paid for. - **Before your API key is delivered:** every charge is refundable in full on request. Email ops@avatar33.com from the address you paid with. - **If we don't deliver your key**, we refund every charge. - After your key is delivered, charges for past periods aren't refunded, except where the law requires. ## Use of the service - Plans are per person. No resale, key sharing, or serving your own product's end users from a flat plan. - "Unlimited" means no per-token billing. Fair-use limits apply as published at [kilowatt.sh/docs/fair-use](https://kilowatt.sh/docs/fair-use). We tell subscribers before limits change. - There is no uptime SLA. Requests may be rate limited at capacity. - Models in the plan may change over time; we'll announce lineup changes on [Models](https://kilowatt.sh/docs/models). - Model outputs are provided as is. You're responsible for how you use them and for complying with each model's license. - We may suspend accounts that abuse the service or break these terms, and refund the unused part of the current period. ## Waitlist - Joining the waitlist is free and creates no obligation for you or us. It doesn't reserve a founding spot. - We may remove signups that are fake, automated abuse, or submitted without authorization. ## Contact ops@avatar33.com --- # Privacy Effective 2026-11-10. Operator: Avatar 8 LLC (ops@avatar33.com). ## What we store - Your email address, and optionally your use case, preferred plan, and a referral note (`ref`) - Whether an agent submitted it (`agent`), the `X-Client` header if sent, and your browser's user-agent string - A salted one-way hash of your IP address, used only for rate limiting and abuse prevention (we don't store the IP itself) - The time you signed up - If you subscribe: your plan, subscription status, and Stripe customer and subscription ids ## Payments Payments are handled by Stripe. Your card details go to Stripe, not to us. Stripe's privacy policy: [stripe.com/privacy](https://stripe.com/privacy). ## What we do with it - Deliver your API key, run your subscription, and email you about Kilowatt. Nothing else. - We don't sell or share your email. ## Where it lives A database hosted by Turso via the Vercel Marketplace; billing records at Stripe. ## Deletion Email ops@avatar33.com from the address you signed up with and we'll delete your entry. Stripe keeps payment records it's legally required to keep.