console only partly convertible
As advertised. Free: 300 credits on a 2-week trial. Pro: 2,000 credits/month. Pro+: 6,000. Ultra: 20,000
What that actually means. A credit is worth $0.02. That is stated outright in the Credit Pack table as a unit price, not inferred -- about EUR 0.0175 or CNY 0.136 at mid-July 2026 rates. It prices top-up credits rather than the included ones, but on that basis every paid tier carries twice its price in credit value. TOKENS: THE FORMULA IS PUBLISHED, ONE CONSTANT IS NOT. Qoder states that "your credit consumption is determined by the number of tokens used per request and the unit price of the model used", and elsewhere that credits follow "the model used and the total token consumption for both input and output". So the relationship is linear and explicit: credits = tokens x model unit price. The per-model unit prices are published too, as the multipliers below. What is missing is a single scalar -- the absolute credits-per-token at any reference multiplier -- and, separately, how many tokens a given request actually consumes. Nothing in 2.7MB of documentation supplies either. So this is not a quota that resists conversion in principle, as several others on this site genuinely do. It is one withheld constant away from being fully computable.
Where to check what is left. in-app usage view
Models covered. Qwen3.7-Max and other Alibaba models, plus BYOK on the free tier
An agentic coding IDE rather than an inference host -- desktop app, JetBrains plugin, CLI, mobile and cloud agents -- so it belongs beside Cursor and Copilot, not beside the model providers. Two things stand out: the free tier allows bring-your-own-key, which most competitors do not, and the site advertises off-peak discounting on Qwen3.7-Max (50% off all day, 80% off-peak), the same peak/off-peak pattern Z.ai applies to quota. Add-on credits expire one month from purchase, and the system spends soonest-expiring credits first. Failed requests are not charged, and basic model calls keep working on a limited daily allowance once credits run out. Published median consumption puts an Editor Ask request at 3-4 credits ($0.06-0.08), an Agent request at 7-12 ($0.14-0.24), and a Quest Experts run at 75 ($1.50). So Pro buys roughly 500 Ask requests or 166 Agent ones a month; Ultra, ten times that. Qoder calls these estimates from its own statistics, not guarantees. THE MODEL YOU PICK CHANGES WHAT YOUR CREDITS ARE WORTH BY 8x. Qoder publishes a credit multiplier per model, and those multipliers do not track what the models cost on the open market. Measured against their blended API prices, one unit of credit buys $5.60 of Qwen3.7-Plus, $5.29 of Kimi-K2.7-Code and $4.43 of Qwen3.7-Max, but only $1.09 of DeepSeek-V4-Pro and $0.68 of GLM-5.2. Alibaba's own Qwen models sit at the favourable end, which is unsurprising -- they cost Qoder nothing to source. A subscriber optimising for value should reach for Qwen3.7-Plus and treat GLM-5.2 as the expensive option, which is the reverse of how those two rank on raw API price. There is also a tier selector on top of the model choice: Lite is free outright (basic reasoning, slower at peak, no multimodal), Efficient runs 0.3x, Auto 1.0x, Performance 1.1x and Ultimate 1.6x -- currently halved to 0.8x for a limited time. Kimi-K2.7-Code carries a Fast toggle that doubles its rate from 0.3x to 0.6x for roughly 6x the speed. AND THE CLOCK MATTERS TOO. Selected models are discounted between 14:00 and 00:00 UTC, weekends and holidays included, affecting credit pricing only and not model quality. Qwen3.7-Max drops from 0.5x to 0.1x off-peak, Qwen3.7-Plus from 0.1x to 0.04x, while GLM-5.2 barely moves (0.6x to 0.5x). Stack that on the per-model spread and the value of a credit swings about 27-fold inside one subscription: off-peak, one unit buys $22.13 of Qwen3.7-Max against $0.82 of GLM-5.2 at open-market prices. Note this runs opposite to Z.ai, which penalises peak hours rather than discounting off-peak. Running out of credits does not stop work: Qoder falls back to basic models with limited messages. But unused plan credits are forfeited on downgrade, and add-on pack credits expire one month after purchase. A note on what is NOT there: the documentation publishes the formula and the relative unit prices but never the absolute rate, nor tokens per request. The practical effect is that you can rank models and time windows precisely against each other -- which is what the figures above do -- while remaining unable to say what any of it costs in tokens.
checked 2026-08-13 · high (read from the rendered pricing page) · source