fix(deepseek): correct V4 Pro/Flash pricing to the published rate card - #6051
Open
settingzbot wants to merge 1 commit into
Open
fix(deepseek): correct V4 Pro/Flash pricing to the published rate card#6051settingzbot wants to merge 1 commit into
settingzbot wants to merge 1 commit into
Conversation
The first-party DeepSeek entries were well below DeepSeek's published prices, most severely on cache reads (6x low for Pro), which dominate agentic traffic. Per https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-09-01), off-peak, USD per 1M tokens: deepseek-v4-pro cache hit 0.022 cache miss 0.66 output 1.98 deepseek-v4-flash cache hit 0.007 cache miss 0.22 output 0.66 Verified against real billing: repricing 30 days of local DeepSeek traffic with these rates reproduces the DeepSeek console's daily spend to the cent ($3.4194 vs $3.42, $0.1920 vs $0.19 on two independent days). The previous rates came out ~3.7x under. Corroborated inside this repo: three other hosts of the same models already price as uniform multiples of the published card - opencode-go at 1.000x, above at 1.100x, aihubmix at 1.048x - while the first-party entry's ratios were non-uniform (0.66 / 0.44 / 0.16), the signature of a stale entry rather than a different pricing basis. Off-peak is used because models.dev has no time-of-day pricing and off-peak covers ~79% of hours; peak (01:00-04:00 and 06:00-10:00 UTC, Mon-Fri) is exactly double. Noted in a header comment on each file. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018EmjRqNQurw9QJhKJGuYuU
Contributor
|
No actionable findings. |
github-actions
Bot
added
the
reviewer: ready
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The first-party
deepseekentries for V4 Pro and V4 Flash are well below DeepSeek's published prices. The gap is worst on cache reads — 6× low for Pro — which dominate agentic traffic, so the error compounds rather than cancels.Published rates
Per https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-09-01), off-peak, USD per 1M tokens:
deepseek-v4-prodeepseek-v4-flashdeepseek-v4-flash-vision-expis priced as Flash, per the existing note in its file.Verified against real billing
I repriced 30 days of my own DeepSeek traffic (token counts from local session databases) with these rates and compared against the DeepSeek console's own daily spend:
To the cent, on two independent days. The current catalog rates come out ~3.7× under.
Corroborated inside this repo
Three other hosts of the same models already price as clean, uniform multiples of the published card, while the first-party entry's ratios are non-uniform — the signature of a stale entry rather than a different pricing basis:
opencode-goV4 ProaboveV4 Pro / FlashaihubmixV4 ProdeepseekV4 Pro (current)The current Pro numbers trace to a May 2026 edit that divided the then-current values by 4; Flash still carries what look like V3-era rates ($0.14 / $0.28).
On peak pricing
DeepSeek charges exactly double during peak (01:00–04:00 and 06:00–10:00 UTC, Mon–Fri). models.dev has no time-of-day pricing concept, so I've used off-peak, which covers ~79% of hours and matches what
opencode-goalready lists. Each file gets a header comment stating the window and the doubling. Happy to switch to peak, or to a blended figure, if maintainers prefer a different convention — worth settling once, since the same question will recur.bun validatepasses; the generated catalog carries the new values.🤖 Generated with Claude Code