Coding Plan (DevPass)

The Coding Plan (DevPass) is a flat monthly plan for coding agents and CLIs — Claude Code, Cursor, Cline, Roo Code, or any OpenAI- or Anthropic-compatible tool. A dedicated ek-dev- key. Unlimited tokens. No credits. Independent of every other Electron Hub plan.

Tiers

	Lite	Turbo
Price	$10/month	$40/month
Token usage	Unlimited	Unlimited
Concurrency slots	2	5
Requests per minute	8 (burst 4)	20 (burst 8)
Priority when busy	Standard	High
Full-speed soft headroom	Generous	6× Lite

Launch seats are capped per tier. If a tier is sold out, join the waitlist on the dashboard. You are notified when a seat opens (first come, first served).

Included models

Series	Model ID	Input	Context · Quant	Notes
Flash · 0.5×	`glm-5.3-flash:dev`	→	1M · FP8	320B-A18B. Native multimodal (text, image, video). Hybrid sparse and linear attention.
Flash · 0.5×	`deepseek-v4-flash:dev`	→	1M · NVFP4 (FP4 MoE + FP8 attn)	Preview Flash. High-throughput coding.
Flash · 0.5×	`deepseek-v4-flash-0731:dev`	→	400K · FP8 + FP4 (MXFP4 MoE)	V4 Flash GA. 13B active. Stronger agent and coding work.
Flash · 0.5×	`mimo-v2.5:dev`	→	1M · FP8	Omnimodal agentic coding.
Standard · 1×	`deepseek-v4.1-flash:dev`	→	1M · FP8 + FP4 (FP4 experts)	552B backbone. 8B prefill / 16B decode active. Native vision.
Standard · 1×	`qwen3.8-27b:dev`	→	262K · BF16	Multimodal agentic coding.
Standard · 1×	`minimax-m2.7:dev`	→	180K · FP8	Agentic productivity and multi-agent work.
Heavy · 2×	`glm-5.3:dev`	→	262K · FP8	744B-A40B. Text-only. Same GLM-5.2 base, post-trained. Stronger complex coding and long-horizon agents. Uses 2 concurrency slots per request.

All models support function calling on /v1/chat/completions, /v1/messages, and /v1/responses. Series weight (0.5× / 1× / 2×) affects soft full-speed headroom (ceil(raw × weight)). It does not stop you from using the plan. Dashboard per-model stats stay raw. Heavy models also hold 2 concurrency slots per request; Flash and Standard models hold 1. See Concurrency slots.

:dev models accept only ek-dev- keys. DevPass keys can call only :dev models. Keep your master key configured if you need the rest of the catalog.

Quickstart

Subscribe on the Dashboard → Console → Coding Plan.
Your ek-dev-... key is provisioned within seconds. It is always retrievable from the Coding Plan tab. Regenerating it immediately invalidates the old key.
Use it as a Bearer token:

curl -X POST "https://api.electronhub.ai/v1/chat/completions" \
  -H "Authorization: Bearer ek-dev-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3:dev",
    "messages": [{"role": "user", "content": "Refactor this function..."}],
    "stream": true
  }'

Fair use

One person, one seat. Lite and Turbo are alternative tiers, not stackable. If two concurrency slots are not enough, upgrade. Do not open another account, email, or payment method to hold a second seat. The ek-dev- key may run on machines you personally use (home, work, laptop, CI, VPN). All of that usage counts against the same seat. The plan is for interactive coding by that one person. These are violations and lead to suspension:

a second seat under any identity you control
sharing the key with anyone else
pooling keys or seats to raise throughput
resale, subletting, or access-as-a-service
a deployed app or backend that serves other people
unattended or multi-user production traffic

We treat extra seats, shared keys, and resale as the same violation. Chargebacks are account-level violations. High-duty-cycle or multi-user traffic belongs on a regular Electron Hub API key billed on credits.

Concurrency slots

Each plan has a fixed number of concurrency slots: 2 on Lite, 5 on Turbo. A request holds its slots until it finishes. Request starts are limited separately (see Requests per minute).

Series	Slots per request	Lite (2 slots)	Turbo (5 slots)
Flash · Standard	1	2 requests	5 requests
Heavy (`glm-5.3:dev`)	2	1 request	2 requests, plus 1 Flash or Standard request

Series can be mixed in any combination that fits the slot total. If the slots a request needs are not free, you get a 429 with Retry-After. Wait for a request to finish, then retry. Coding agents handle this automatically.To keep parallel work moving on Lite, route subagents and background tasks to a Flash model and reserve the Heavy model for the main session. While the plan is temporarily reduced to one slot, a Heavy request uses that single slot.

Full-speed soft headroom (daily)

Each plan includes full-speed headroom that resets at 21:00 UTC. Going past it never blocks you. Requests continue with reduced concurrency and paced admission (consumption-scaled spacing, up to about a minute when far into overage) until the reset. The dashboard shows how close you are to full-speed headroom.

Full-speed soft headroom (weekly)

A rolling weekly soft headroom sits above the daily one. Past it, the same slow lane applies (lower concurrency and paced admission). This is not a hard stop. It keeps sustained 24/7 grinders from crowding out interactive work across the week.

Requests per minute

Request starts refill continuously at 8 per minute on Lite and 20 per minute on Turbo. Up to 4 (Lite) or 8 (Turbo) can start back-to-back after a quiet moment. Every request counts as one start, Heavy included. In the slow lane (past full-speed headroom, or in low-interactivity mode) it is 3 per minute (burst 2).Past the limit you get a 429 with an exact Retry-After (Lite: about 8 seconds per start). X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset report this budget on every response. These 429s never count toward the temporary pause.

Temporary pause after concurrency hammering

Repeatedly hitting the parallel-slot limit (hundreds of concurrency 429s in a usage day) can pause admission for a few hours. While paused, requests return 403. Your key stays valid. Service resumes automatically at the time shown on the dashboard. Regenerating the key clears an automatic concurrency pause.

Low-interactivity mode

Plans are sized for a person coding, even intensively, all day. When an account shows a usage pattern far beyond typical interactive work, the plan enters low-interactivity mode for about 24 hours:

Under light load, behavior is unchanged.
When the platform is busy, requests may be queued behind interactive sessions and may need a retry.
Nothing is blocked. Your plan, models, and answer quality stay the same.
Interactive mode resumes automatically at the time shown on the Coding Plan tab. No action is required.

Low-interactivity mode is a temporary fairness control so interactive subscribers keep responsive access during busy periods. It is not a suspension. If this surprises you, check for a runaway automation first, then ask on Discord.

Errors

Status	Meaning	What to do
`429`	Not enough free concurrency slots (Heavy requests need 2), or the per-minute request limit was reached	Honor `Retry-After`, then retry
`404` model error	Model not in the Coding Plan list	Use a `:dev` model from the table above
`403`	Temporarily paused after concurrency hammering	Wait for the resume time on the dashboard, or regenerate your key to clear the pause
`403`	Subscription suspended or ended	Check payment, or resubscribe. A second seat or shared use can also cause this.

Terms

Monthly billing via Creem, auto-renewing, non-refundable. Cancel any time. Your key stays active until the period ends.
Fair use is binding. Extra seats, shared keys, pooling, and resale are suspended. Chargebacks are account-level violations.

FAQ

Is token usage really unlimited?

Yes. There is no hard token cap and no per-token charge. Soft full-speed headroom only affects concurrency, request rate, and latency when you are far past typical interactive use. Requests are never blocked for burning tokens.

I bought during beta. Do I keep the old price?

Yes. Founding and beta subscribers who purchased at the launch rate keep that price for as long as the subscription stays active. If you cancel and subscribe again later, the current public price applies.

Can I buy a second Coding Plan seat?

No. See Fair use. One person, one seat. Upgrade Lite to Turbo if you need more parallel slots.

Can I upgrade Lite to Turbo?

Yes. Change plans in the Creem billing portal. Your existing key picks up Turbo limits automatically.

What happens when my subscription ends?

Your ek-dev- key is deactivated. Resubscribe any time to restore it.

What does the Low-interactivity mode banner mean?

See Low-interactivity mode. Requests remain available. Under load they may queue behind interactive sessions until interactive mode resumes automatically.

Getting Started

API Reference

Guides

Examples

Billing

Tiers

Included models

Quickstart

Fair use

Low-interactivity mode

Errors

Terms

FAQ

​Tiers

​Included models

​Quickstart

​Fair use

​Low-interactivity mode

​Errors

​Terms

​FAQ

Tiers

Included models

Quickstart

Fair use

Low-interactivity mode

Errors

Terms

FAQ