A dated source snapshot
Prices and access conditions were checked against official pages on 30 September 2026. Each row links its source. These snapshots are not a live billing feed. See sources and held-out candidates.
The same usage on both sides
Default input:output is 5:1, an editable example, not a universal workload. Exact tokens price identical counts for both models. Include billed reasoning in output. Different tokenizers and response lengths mean this is not a guaranteed cost for the same task.
Decimal arithmetic is used for costs. Usage details accepts cached reads and known extra fees. Unknown bill composition cannot be safely inverted into usage. Missing prices are never treated as free. Taxes, cache writes/storage, tools, input materials and retries are excluded unless explicitly accounted for.
The DeepSeek rate-window control applies a complete peak or off-peak tariff to both sides. It does not average a mixed monthly schedule. Price bracket selectors apply one provider bracket to each model’s usage. Monthly token totals do not determine per-request brackets. Compare mixed brackets in separate groups. OpenAI uses the provider’s Short/Long labels; Grok switches at 200K prompt tokens and MiniMax M3 above 512K input tokens. Media rates retain their native units and are not processed by the text calculator.
Display currency
USD is the original quote. Other currencies use ECB reference rates dated2026-09-30, crossed from the euro base. Conversion changes displayed amounts, not usage, purchasing eligibility or savings percentage. These informational rates exclude card spreads and exchange fees and are not transaction quotes. ECB source ↗
Three subscription highlights
Lowest monthly fee compares regular monthly renewals among the five listed plans, excluding introductory and annual-payment discounts. It is not a whole-market or model-quality award.
Most tokens and best full-use value compare the three GLM plans with published credit conversions, using your selected GLM model and workload. MiniMax remains excluded from these two rankings because a numerical quota is not confirmed. One plan may lead both categories. This is a scenario comparison, not a whole-market or quality ranking.
Credits equal (uncached input × input multiplier + cached input × cache multiplier + output × output multiplier) / 10,000. GLM-5.3 uses 6.9 / 1.7 / 24; Flash uses 2.3 / 0.56 / 8. Regular off-peak halves credit consumption. The default is GLM-5.3, input:output 5:1, zero cached input, regular peak rates and full usage. Temporary discount and unlimited-night promotions are excluded. See the official credit rules.
The 28-day example uses four complete weekly windows, with five aggregate usage blocks each week, at least five hours apart, and no opening usage. Each block may contain many requests. Both rolling five-hour and weekly limits are simulated. It does not establish a monthly maximum; an actual billing month, reset anchor, daily demand, rate changes and other tool consumption may give a different result. The same workload is priced against the matching official GLM API tariff from our dated snapshot.
Break-even usage = one regular monthly fee / API equivalent at full usage across those four quota weeks. It is the share of this example’s token volume needed for API spending to equal the fee. Above 100% means the plan cannot break even within this example. Potential saving uses your usage setting; unused tokens have no cash value. Cache writes/storage, MCP tools, taxes and other charges are excluded. The two ranking winners use full capacity even when your usage setting is lower.
Editorial sources
The English briefs are original summaries of linked primary sources. Their dates mark editorial review, not necessarily product release. Vendor descriptions and availability claims are not independent performance tests or endorsements.
Privacy
This site has no accounts, analytics, email collection or runtime AI service. Calculator inputs stay in your browser.