No token meter. Run your coding agents all day on open-weight DeepSeek V4.1 Flash and GLM-5.3 for one flat price, with a smart router and an open-source agent included.*
*Unlimited means no per-token billing. Per-minute and daily fair-use limits apply; see fine print.
Live count of real signups. Never padded.
Built for people whose agents burn through tokens: Claude Code, Codex-style CLIs, OpenCode, your own loops. Point them at one endpoint and stop watching a meter.
One monthly price. No per-token bills, no surprise invoices. Cached input never counts against your limits.
OpenAI- and Anthropic-compatible endpoints. Two environment variables move Claude Code over.
Easy requests go to DeepSeek V4.1 Flash, hard ones to GLM-5.3. Pin a model any time; a response header explains each route.
DeepSeek V4.1 Flash (MIT license) and GLM-5.3. No closed-model lock-in.
A hosted open-source agent (browser, terminal, memory, routines) running on the same flat plan, rolling out to founding members.
Agents can join the waitlist with one JSON POST. Spec in openapi.json and llms.txt.
Founding pricing for the first 100 subscribers. Nothing is charged today: joining the waitlist is free and holds your spot.
Same idea: one flat price, no token meter. Less than a fifth of the price.
| Kilowatt Pro | Morph Ultra | |
|---|---|---|
| Price | $39 / month (Lite $19) | $200 / month |
| Unmetered models | DeepSeek V4.1 Flash + GLM-5.3 | Kimi K3.1 + DeepSeek V4.1 Flash |
| Model router | Included | morph-router sold separately at $0.005 / request |
| API formats | OpenAI + Anthropic | OpenAI + Anthropic |
| Cached input | Exempt from limits | Exempt from limits |
| Seats | Founding 100, then waitlist | 1,000 subscriptions |
| Status | Early access waitlist | Live |
Morph figures from morphllm.com/products/ultra and morphllm.com/pricing, read Nov 10, 2026. Morph's per-minute limits are higher than ours; see the fine print. Morph and Morph Ultra are trademarks of their owner; Kilowatt is not affiliated with Morph.
We haven't published our own benchmarks yet. These are the vendors' self-reported scores, at max reasoning effort, from their model cards. Your results through any provider can differ.
| Benchmark (vendor-reported) | DeepSeek V4.1 Flash | GLM-5.3 |
|---|---|---|
| Terminal-Bench 2.1 (pass@1) | 90.6 | 88.2 |
| DeepSWE v1.1 (resolved) | 74.2 | 66.9 |
| NL2Repo | 64.0 | 58.0 |
Sources: DeepSeek-V4.1-Flash model card (DeepSeek Harness / mini-SWE harness) and GLM-5.3 model card (Claude Code / mini-swe-agent harness). Harnesses differ, so the columns are not a head-to-head comparison.
When your founding access opens you get an API key that works with Claude Code and any OpenAI SDK. Until then, humans use the form below and agents use the API.
# Preview: how Claude Code will connect (access not open yet) export ANTHROPIC_BASE_URL="https://api.kilowatt.sh" export ANTHROPIC_AUTH_TOKEN="$KILOWATT_API_KEY" export ANTHROPIC_MODEL="glm-5.3" export ANTHROPIC_SMALL_FAST_MODEL="deepseek-v4.1-flash" claude
# Works today: join the waitlist from an agent curl -s https://kilowatt.sh/api/waitlist \ -H 'content-type: application/json' \ -d '{"email":"you@example.com", "use_case":"coding agent", "agent":true}' # → {"ok":true,"position":12,"founding":true,...}
Free. No card. The first 100 signups are the founding cohort at founding pricing; everyone after that joins the waitlist in order.
"Unlimited" is the honest kind: no token meter, but fair-use speed limits so nobody crowds out everyone else.
There's no per-token billing and you're never charged for usage. Each subscription has fair-use limits on how fast and how much you use per day:
When you hit a limit you get an HTTP 429 with Retry-After; OpenAI and Anthropic SDKs retry automatically. Limits may be adjusted as capacity grows; subscribers will be told about changes.
Morph publishes higher per-minute limits (for DeepSeek V4.1 Flash: 2,103,237 uncached input and 76,559 output tokens/min per subscription). Ours are sized to what one developer's agents typically use, which is how we can charge $39 instead of $200. If you need more throughput, tell us your use case on the form.
Not yet. This is an early-access waitlist. The founding 100 get access first, in signup order. We'll email you when your spot opens.
No. Joining is free and needs no card. Pricing applies only if you choose to subscribe when access opens.
There is no uptime SLA during early access. Requests can be rate limited at capacity. We'll publish live speed and error rates once we're serving traffic.
DeepSeek V4.1 Flash (MIT license) and GLM-5.3, both open-weight. The router picks between them unless you pin one. We may add models over time.
A hosted open-source agent with a browser, terminal, files, memory, and scheduled routines, running on the same flat plan. It's rolling out to founding members after API access.
No. Plans are per person. Sharing keys outside your account or serving your product's end users from a flat plan isn't allowed.
Kilowatt is built by Avatar 8 LLC, an independent company. We're not owned by a model lab, cloud, or payments company.