// AI Platform Drift
DeepSeek V4 Flash pricing dropped 90% on completion tokens
OpenRouter model catalogue · published 2026-09-30 22:00 UTC
CORRECTED 2026-10-02 17:58 UTC — [CORRECTION · WHY IT MATTERS] Original: “Total cost per request drops 89-90% across all workload mixes. A workload generating ten completion tokens for every prompt token now costs $0.0014/M tokens instead of $0.0126/M tokens. Existing budget forecasts are now overstated by roughly 10x.” Fact: Completion pricing falls 89.5% and prompt pricing falls 17.5%; a universal 89–90% blended decrease is not supported. For 1M prompt tokens plus 10M completion tokens, captured cost is $12.5120 before and $1.3167 after (89.5% lower). Across all nonnegative token mixes, the decrease lies between 17.5% and 89.5%.
[CORRECTION · WHAT TO DO] Original: “you have 90% of it available for other work or can reduce the allocation.” Fact: Any available budget depends on actual token volumes and mix; recompute both components instead of assuming a 90% saving.
Stood 43h 58m · cause arithmetic · every correction ↗
Captured arithmetic only; vendor rate validity, applicability and effective conditions remain unverified.
WHAT CHANGEDOpenRouter reduced pricing for deepseek/deepseek-v4-flash-latest: completion tokens from $1.25/M to $0.13/M (-89.5%), prompt tokens from $0.012/M to $0.0099/M (-17.5%). Change is live as of 30 September 2026.
WHY IT MATTERSTotal cost per request drops 89-90% across all workload mixes. A workload generating ten completion tokens for every prompt token now costs $0.0014/M tokens instead of $0.0126/M tokens. Existing budget forecasts are now overstated by roughly 10x.
WHAT TO DOReforecast monthly spend for any service using this model and notify finance or whoever owns the AI budget. If you allocated budget assuming the old rate, you have 90% of it available for other work or can reduce the allocation.
Model-reported confidence 95% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
id: ~deepseek/deepseek-v4-flash-latest name: DeepSeek: DeepSeek V4 Flash Latest -pricing.prompt: 0.000000012 -pricing.completion: 0.00000125 +pricing.prompt: 0.0000000099 +pricing.completion: 0.00000013068 context_length: 1048576 top_provider.max_completion_tokens: 943718
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.95, needed 0.5 |
|---|---|
| triage reasoning | Significant price reductions on both prompt and completion tokens for an active model. |
| signals | prompt pricing decreased ~17% · completion pricing decreased ~90% · pricing change on active model |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |