plans verified 2026-08-29

AI coding subscription: flat-rate LLM plans for your coding agent

Standard Compute is an LLM subscription for coding agents — OpenCode, Cline, Aider, Codex CLI, Kilo, Claude Code and always-on agents — with every frontier model behind one OpenAI-compatible endpoint, from $39/month flat. No token meter, no 5-hour windows, no weekly caps.

Below: how it compares, honestly, with every coding plan researchers actually shortlist — including the ones that beat us on price and the cases where they're the right pick.

PlanPriceModelsLimit modelThe catch
Standard Compute (this site)$39 / $89 / $249 / $499 per moFrontier + open (GPT-5.6, Claude 5, Gemini, DeepSeek, Qwen, Kimi, GLM) via smart routingNo token meter, no session/weekly windows — fixed monthly compute budget; sustained extremes paced smoothly, never cut off mid-taskRouting picks the model — you don't pin one. Budget is generous, not infinite.
Claude Max$100 / $200 per mo (Pro $20)Claude only (incl. Claude Code)Fixed price with metered usage: ~5-hour session windows plus weekly caps; Max raises the caps ~5-20x over Pro.Single vendor, and heavy agentic use hits the windows.
OpenCode Go$10/mo ($5 first month)12 open-weight (DeepSeek V4, Qwen 3.6, GLM 5.2, MiniMax M3…)Fixed price buying ~6x its value: ~$60/mo of usage-equivalent, metered at model rates.A capped value multiplier with tight windows: $12 per 5 hours, $30 per week, $60 per month. Open-weight models only.
MiniMax Token Plan$22 / $55 / $132 per mo (per docs; MiniMax's own M3 blog says $20/$50/$120)MiniMax M3 (1M context, image/video input, computer use) + M2.7, plus image/speech on a shared quotaFixed price with per-tier 5-hour rolling and weekly windows, no carry-over — exact per-tier numbers aren't published.MiniMax's docs page prices the tiers at $22/$55/$132 while its June 2026 M3 launch blog says $20/$50/$120 — confirm at checkout. Window sizes are unpublished; H3 and voice-design/voice-clone models are excluded.
GLM Coding Plan (Z.ai)$18 (Lite) / ~$80 / ~$168 per mo (quarterly/yearly discounts)GLM only (GLM-5.3 flagship + GLM-5.3-Flash; requests for 5.2/5.1 auto-route to 5.3, 4.7 to 5.3-Flash)Fixed price with dual-window credit quotas: 2,000 / 12,000 / 28,000 credits per 5 hours and 10,000 / 60,000 / 140,000 per week (Lite/Pro/Max).Credits, not prompt counts, since the 2026 revamp — burn varies by model, so capacity is less legible than a request cap. GLM models only; only the $18 Lite price is pinned on official docs, so confirm Pro/Max at checkout.
Qwen Coding Plan (Alibaba)$50/mo (Pro — the only tier still open to new subscriptions)Multi-vendor despite the name: Qwen 3.x line (incl. vision + coder models) plus Kimi K2.5, GLM-5, GLM-4.7, MiniMax-M2.5Fixed price with three simultaneous, fully published windows: 6,000 requests per rolling 5 hours, 45,000/week, 90,000/month.The ~$10 Lite tier was discontinued for new subscriptions in March 2026 — and no 'QwenCloud Standard $18' tier exists, despite AI-generated comparison tables claiming one. Plan keys don't work on the general Model Studio API.
Synthetic$1/day or $30/mo8 open-weight (Kimi-K3, GLM-5.2, GLM-5.3-Flash, Nemotron-3-Super-120B, gpt-oss-120b, Qwen3.8…)Fixed price with a request quota — 500 requests per 5 hours — and no token metering at all.1 concurrent request per model (extra concurrency sold separately) hurts parallel-agent setups; two lineup models are flagged Beta; OpenAI-compatible only.
Chutes$3 / $10 / $20 per mo13+ open models (GLM-5, Kimi K2.5, Qwen 3.5, MiniMax M2.5… — flagships on Plus/Pro only)Fixed price with a bundled daily quota, capped since Feb 2026 at 5x the equivalent pay-as-you-go value — possibly enforced over shorter rolling windows.Flagship models were pulled from the $3 Base tier; exact per-tier request numbers aren't on the official page (launch-era figures said ~2,000/day on Plus); decentralized infra and a volatile model lineup.
Cline Pass$9.99/mo ($4.99 first month)11 open-weight (GLM 5.2, Kimi K3, DeepSeek V4 Pro, Qwen3.7-Max…)Fixed price buying 2-5x the standard API rate limits on a curated open-weight lineup — quota is rate-limit headroom, not a published dollar amount.Quotas stated as rate-limit multiples (2-5x), not dollar windows, so what you actually get is opaque; open-weight only; 'additional processing fee may apply'.
DevPass$29 / $79 / $179 per mo200+ models via gatewayFixed price buying ~3x its value in usage metered at provider rates (Lite ≈ $87 of usage, Max ≈ $537).A capped value multiplier: consume the included value and the month is done.

All plan facts re-verified 2026-08-29. Full field including 14 more budget plans: the complete coding-plans comparison →

connect your agent in 2 minutes
base_url = https://api.stdcmpt.com/v1 · model = standardcompute · Claude Code: ANTHROPIC_BASE_URL = https://api.stdcmpt.com

FAQ

What is the best AI coding subscription in 2026?
It depends on your usage shape. Claude Max if Claude Code within session windows covers you; OpenCode Go or the Chinese token plans (MiniMax, GLM, Qwen) for the cheapest open-weight quota; Standard Compute if you want one subscription with frontier AND open models, no 5-hour or weekly windows, behind the agent you already use. The honest comparison table on this page states each plan's real limit model.
How is Standard Compute's coding subscription different from token plans?
Token/credit plans (OpenCode Go, MiniMax, GLM, Synthetic, Chutes) grant a metered quota inside rolling windows — cheap until a heavy sprint exhausts the window mid-task. Standard Compute has no token meter and no windows: sustained heavy load is paced smoothly instead of stopped, and the monthly price is the whole bill. Frontier models (GPT-5.6, Claude 5, Gemini) are included via smart routing, not open-weights-only.
Which coding agents work with it?
Any OpenAI-compatible agent: OpenCode, Cline, Aider, Roo, Kilo, Continue, plus always-on agents like OpenClaw and Hermes. Claude Code connects via the Anthropic Messages API (ANTHROPIC_BASE_URL swap). Setup is a base URL, a key, and model 'standardcompute'.
Is it really unlimited?
No coding subscription is truly unlimited, and we say so plainly: each plan includes a generous fixed monthly compute budget — the difference is there's no token meter counting down and no session windows interrupting work. Extreme sustained load is paced smoothly rather than cut off. New accounts get free starter credit to verify the experience before paying.

Per-agent setup guides: OpenCode, Claude Code, Cline, Aider · how flat-rate works · for teams