DeepSeek V4.1 Flash GOAT credits are charged ~3.03× raw usage cost despite advertised $60 allowance

[bitemiaow]

Summary

DeepSeek V4.1 Flash usage on my GOAT plan appears to consume plan credits at a much higher multiplier than implied by the advertised $60/month per-model allowance.

I verified the raw request costs, cache pricing, peak/off-peak pricing, aggregated chart data, and billing credit ledger through Command Code's own internal API endpoints.

Plan / model

• Plan: GOAT
• Monthly credits granted: $70
• Model: deepseek/deepseek-v4.1-flash
• Usage mode: Provider API
• Advertised DeepSeek V4.1 Flash allowance: $60/month

Reproduction / evidence

For the same snapshot of 244 completed requests, /internal/usage/summary returned:

• totalCount: 244
• totalTokensIn: 54,221,895
• totalTokensOut: 213,734
• totalCost / totalCredits: $2.056627545

I then summed every matching bucket from /internal/usage/charts.

The buckets contain the exact same totals:

• requests: 244
• input tokens: 54,221,895
• output tokens: 213,734

But the sum of totalCost from those buckets is only:

$0.679399746

Therefore:

2.056627545 / 0.679399746 = 3.0271×

So approximately 3.03 plan credits are being deducted per $1 of raw model usage.

Raw request pricing is correct

Individual /internal/usage entries correctly match the advertised V4.1 Flash token rates.

For example, during off-peak pricing:

• fresh input: $0.15 / M
• cached input: $0.003 / M
• output: $0.60 / M

The inputCost, cacheCost, outputCost, and totalCost fields reconcile exactly token-by-token.

Cache hit rate in the sampled requests is also ~99%+, so the discrepancy is not caused by cache misses.

Expected behavior

If the advertised $60/month V4.1 Flash allowance on a $70 GOAT credit pool is applied using the proportional allowance mechanism, the multiplier should be approximately:

70 / 60 = 1.1667×

For $0.679399746 of raw model usage, that would imply approximately:

$0.792633 credits

Instead, $2.056628 credits were deducted.

The observed ~3.03× multiplier corresponds to an effective per-model allowance of only about:

70 / 3.0271 ≈ $23.12

Question

Is the advertised $60 DeepSeek V4.1 Flash GOAT allowance currently applied to Provider API usage?

If yes, why is the billing ledger applying an effective ~3.03× multiplier?

Could you please check whether the V4.1 Flash promotional allowance has been correctly deployed to the billing backend?

Thanks.

[ahmadbilaldev]

Charges are correctly setup. Please reach out to support@commandcode.ai for billing related queries for a deeper look.