AI API pricing comparison: GPT, Claude and Gemini — live group snapshot
Stop comparing blog screenshots. The MonkeyAI /pricing/ page re-renders on every visit from channels + channel_model_pricing + groups + channel_groups, so the row you see after login is the row to budget against. This guide shows the rows visible on 2026-08-08 and how to read them.
Why price-comparison reading order matters
Direct answer: compare the live prices your account can actually see, not the numbers on a third-party blog. The MonkeyAI pricing page is built from channels + channel_model_pricing + groups + channel_groups and re-renders on every visit, so the row you see after login is the only row that matters for buying decisions.
https://monkeytheme.com/pricing/ on 2026-08-08. The page reported 2 product families, 4 visible pricing groups, and 9 visible models. Numbers are quoted as ¥ per 1M tokens for input, output and cache reads.What a "visible price" really is
A visible price is the result of three multipliers stacked on top of the model's official USD reference rate. The page you see at /pricing/ shows the live result; guessing it from memory or from another vendor's blog is what causes pricing mismatches.
| Layer | What it controls | Why it matters for comparison |
|---|---|---|
channel_model_pricing | Base USD rate per token for input / output / cache read | Defines the official reference price in the row |
groups.rate_multiplier | Per-group multiplier on the base rate | Different groups see different prices for the same model |
groups.supported_model_scopes | Which models a group can see at all | Some models are not available to every group |
When two prices disagree, the difference is usually in one of these three layers — not in a hidden discount code.
Codex / OpenAI pricing — what is live today
Four GPT/Codex models are currently visible under the Codex (Enterprise) group with a 1.0500x multiplier. Every row is roughly 85.0% below the official reference; the cache-read column is the cheapest line because it is reused context that the upstream does not have to re-generate.
| Model | Group | Input (¥/1M) | Output (¥/1M) | Cache read (¥/1M) |
|---|---|---|---|---|
gpt-5.3-codex | Codex (Enterprise) · 1.0500x | ¥1.84 | ¥14.70 | ¥0.19 |
gpt-5.4 | Codex (Enterprise) · 1.0500x | ¥2.63 | ¥15.75 | ¥0.26 |
gpt-5.4-mini | Codex (Enterprise) · 1.0500x | ¥0.79 | ¥4.73 | ¥0.08 |
gpt-5.5 | Codex (Enterprise) · 1.0500x | ¥5.25 | ¥31.50 | ¥0.53 |
If you are picking a default for an agent loop, gpt-5.4-mini has the lowest absolute cache-read price (¥0.08 / 1M) and the largest discount percentage. gpt-5.5 has the highest output price in the family but is also the strongest on long-context tasks; do not switch to it for cheap completions.
Claude pricing — three groups, one model family
The Claude family is sold under three groups, each with a different multiplier. All three groups expose the same five models, but the per-token price differs because the multiplier is applied to the same base USD rate.
| Model | Input (¥/1M) | Output (¥/1M) | Cache read (¥/1M) |
|---|---|---|---|
claude-haiku-4-5-20251001 | ¥1.05 | ¥5.25 | ¥0.11 |
claude-opus-4-6 | ¥4.46 | ¥18.11 | ¥0.37 |
claude-opus-4-7 | ¥5.25 | ¥26.25 | ¥0.53 |
claude-opus-4-8 | ¥5.25 | ¥26.25 | ¥0.53 |
claude-sonnet-4-6 | ¥3.15 | ¥15.75 | ¥0.32 |
| Model | Input (¥/1M) | Output (¥/1M) | Cache read (¥/1M) |
|---|---|---|---|
claude-haiku-4-5-20251001 | ¥1.50 | ¥7.50 | ¥0.15 |
claude-opus-4-6 | ¥6.38 | ¥25.88 | ¥0.53 |
claude-opus-4-7 | ¥7.50 | ¥37.50 | ¥0.75 |
claude-opus-4-8 | ¥7.50 | ¥37.50 | ¥0.75 |
claude-sonnet-4-6 | ¥4.50 | ¥22.50 | ¥0.45 |
| Model | Input (¥/1M) | Output (¥/1M) | Cache read (¥/1M) |
|---|---|---|---|
claude-haiku-4-5-20251001 | ¥3.75 | ¥18.75 | ¥0.38 |
claude-opus-4-6 | ¥15.94 | ¥64.69 | ¥1.31 |
claude-opus-4-7 | ¥18.75 | ¥93.75 | ¥1.88 |
claude-opus-4-8 | ¥18.75 | ¥93.75 | ¥1.88 |
claude-sonnet-4-6 | ¥11.25 | ¥56.25 | ¥1.13 |
The same claude-opus-4-7 model costs ¥5.25 / 1M input on Claude lite and ¥18.75 / 1M input on Claude Max. The base rate is identical; only the multiplier changes.
Claude lite (Special), Claude Plus (Premium) and Claude Max (Enterprise) are stable identifiers; the Chinese alt labels (特价 / 精品 / 企业级) change with the UI build.How the "% below official" is calculated
The pricing page shows two columns per row: the live price and the official reference price. The percentage badge is computed as (official - live) / official × 100.
# Worked example, gpt-5.4-mini input official_yuan = 5.25 live_yuan = 0.79 discount_pct = (official_yuan - live_yuan) / official_yuan * 100 # => 84.95...% rounded to 85.0%
Because the badge is recomputed on every visit, a small live-price change can keep the badge at the same integer percent even when the live row has moved by ¥0.01. Treat the percentage as a stable label, not a precise number.
What to compare before picking a model
| Factor | Why it changes the buying decision |
|---|---|
| Input vs output ratio | Code generation shifts cost to output; long-context RAG shifts cost to input |
| Cache read reuse | Long system prompts with cache hits make cache-read the dominant line |
| Group multiplier | Same model, different group → 3-7x price difference for Claude family |
| Task success rate | A cheaper model that needs two retries can cost more than one premium call |
| Latency / timeout behavior | Slow upstream can trigger 504 and force a paid retry |
The cheapest visible model is rarely the right default. The right default is the cheapest model that handles the task at the first attempt.
What this article does not cover
- Image / per-request billing. The current visible models are all token-priced. If a future image model is added, the row would use a different billing_mode.
- Subscription or group membership changes. Pricing here reflects the rows visible at capture time, not what your specific account is on.
- Billing invoices, refunds, or top-up logic. See the live
/billingview for that. - Per-token USD math. The pricing page hides the USD rate and only shows
¥ / 1M tokens; if you need the raw USD multiplier, query the admin API directly.
Related guides and references
Frequently asked questions
Does MonkeyAI show GPT, Claude and Gemini prices in the same place?
GPT and Claude are shown on the live /pricing/ page under product-family tabs. Gemini was not visible to the captured account on 2026-08-08; only models with an active channel + visible model scope appear.
Are the official reference prices the same as OpenAI and Anthropic list prices?
They follow the official USD list rate converted at the page's display exchange rate. They are not affected by promotions or contract pricing at the upstream vendor.
Why is the same Claude model cheaper in Claude lite than Claude Max?
Each group applies its own rate_multiplier on the same base USD rate. Claude lite is 1.0500x, Claude Max is 3.7500x, so the same model is roughly 3.6x more expensive on Max.
Should I switch groups to save money?
Group eligibility is set in the admin panel, not by the customer. What you can pick is which model within your visible group to call.
Why does the live price page show two numbers per row?
The first number is the live price after the group multiplier; the second is the official reference price. The badge is the percentage difference.
Will the live prices change without me changing anything?
Yes, if the upstream vendor adjusts the USD reference rate or if the admin panel updates a group multiplier, the live row will change on the next page load.