Flat-Rate AI Coding Plans, Compared Honestly

Every plan on this page calls itself flat-rate. They mean three different things: a fixed price with metered usage, a fixed price buying a capped multiple of usage, or genuinely unlimited tokens. We sell one of these — so every entry below, including ours, has its catch stated in plain text. Facts verified 2026-07-17.

PlanPriceModelsWhat 'flat' means hereThe catch
Claude Pro / Max$20 / $100 / $200 per moClaude 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 — 'usage limit reached' is one of 2026's most-searched errors.
ChatGPT Plus / Pro$20 / $200 per moOpenAI only (incl. Codex)Fixed price with rolling usage windows shared across Codex local and cloud tasks.Single vendor; parallel Codex tasks burn the window fast.
GitHub Copilot Pro / Pro+$10 / $39 per moCurated multi-modelFixed price with monthly premium-request allowances.Premium-request limits and add-on charges complicate the 'simple' price.
Cursor Pro / Ultra$20 / $200 per moCurated multi-model, in-editorFixed price buying roughly its own value in usage (Pro ≈ break-even, Ultra ≈ 2x).A value meter, not unlimited — and it lives inside one editor.
DevPass (LLM Gateway)$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: run through the included value and you're done for the month. Genuinely good multiple — not unlimited.
Featherless$25 / $100 / $200 per moOpen-weight only (DeepSeek, GLM, Kimi…)Unlimited tokens at a fixed price.No frontier proprietary models, concurrency caps (4-8 connections), and context caps (32K on the $25 tier, 256K above).
Standard ComputeTHAT'S US$9 / $39 / $99 / $399 per moFrontier pool, auto-routed (one model id)Unlimited tokens at a fixed price; tiers differ by execution speed, not usage.You don't pin model versions (smart routing picks per request), sustained extreme load is paced rather than instant, and plans include one API key. That pacing is the mechanism that makes unlimited economically real.

Prices and terms from each provider's public pages and positioning as of 2026-07-17. Spot an error? Tell us and we'll fix it — the point of this page is being right.

The three meanings of "flat-rate"

1. Provider plans: flat price, metered usage, one vendor

Claude Max and ChatGPT Pro bundle excellent agents with predictable prices — and rolling usage windows that agentic workloads exhaust. If you've searched 'usage limit reached', you know the ceiling is real. Best when you live inside one vendor's ecosystem and your volume fits the windows.

2. Value-multiplier gateways: flat price, capped value

DevPass and Cursor Ultra sell a discount, not unlimited usage: your subscription buys roughly 2-3x its price in provider-rate usage, then stops. That's honest value if your monthly usage is predictable and fits the cap — and a mid-month wall if your agent has a big week.

3. Unlimited flat-rate: no token meter at all

The only category where 'how much did my agent use this month?' stops being a question. The economics require a different give: Featherless limits you to open-weight models with concurrency and context caps; Standard Compute serves frontier models but paces sustained extreme load instead of rejecting or billing it. Nothing unlimited is free of trade-offs — the difference is which trade-off fits your work.

Every head-to-head, compared

Deciding between two specific plans? Each matchup has its own honest verdict.

ChatGPT Plus / Pro vs Claude MaxClaude Max vs GitHub Copilot Pro / Pro+Claude Max vs Cursor Pro / UltraClaude Max vs DevPassClaude Max vs FeatherlessChatGPT Plus / Pro vs GitHub Copilot Pro / Pro+ChatGPT Plus / Pro vs Cursor Pro / UltraChatGPT Plus / Pro vs DevPassChatGPT Plus / Pro vs FeatherlessCursor Pro / Ultra vs GitHub Copilot Pro / Pro+DevPass vs GitHub Copilot Pro / Pro+Featherless vs GitHub Copilot Pro / Pro+Cursor Pro / Ultra vs DevPassCursor Pro / Ultra vs FeatherlessDevPass vs Featherless

Flat-rate plans — common questions

What is the best flat-rate AI coding plan in 2026?

It depends on which 'flat' you need. For living inside one vendor: Claude Max ($100-$200) or ChatGPT Pro ($200). For a predictable discount on multi-model usage: DevPass (~3x value, capped) . For genuinely unlimited tokens: Featherless (open-weight models, from $25) or Standard Compute (frontier models with smart routing, from $9, paced under sustained heavy load).

Is any AI coding plan actually unlimited?

Most 'flat-rate' plans are flat in price, not usage: provider plans have session windows and weekly caps, and value-multiplier gateways stop when you've consumed your included value. Genuinely unlimited-token plans exist (Featherless, Standard Compute) and each pays for it with a different trade-off — model selection limits or speed pacing — because truly uncapped, full-speed frontier usage at a fixed price is not economically possible for anyone.

Is DevPass unlimited?

No — and it doesn't claim to be. DevPass sells a value multiple: each tier includes roughly 3x its price in usage metered at provider rates (about $87 of usage on the $29 Lite tier, per its own positioning, verified 2026-07-17). That's a genuine discount with a monthly cap, which suits predictable usage and stops a heavy month mid-stride.

Claude Max vs a flat-rate gateway — which should I pick?

Claude Max if Claude Code specifically is your daily driver and your usage fits its windows — it remains the quality benchmark. A gateway when the windows keep interrupting you or you want multi-model freedom. Many heavy users run both: Max for judgment-heavy work, an unmetered endpoint for volume.

Compute your own workload →The per-token vs flat math →Why metered plans hit walls →