// AI Platform Drift
DeepSeek V3 prompt pricing dropped 19.6%, completion up 2.3%
OpenRouter model catalogue · published 2026-09-27 16:33 UTC
CORRECTED 2026-10-02 17:58 UTC — [CORRECTION · headline] Original: “completion up 2.3%” Fact: Completion price rises 15.6%, not 2.3%.
[CORRECTION · WHAT CHANGED] Original: “completion tokens from $16,384/M to $16,000/M (-2.3%).” Fact: Completion pricing is $0.89/M to $1.0287/M (+15.6%); 16,384 to 16,000 are output-token limits.
[CORRECTION · WHY IT MATTERS] Original: “Total cost per request decreases 2.3% across all workload mixes.” Fact: Net cost depends on the token mix: prompt pricing falls 19.6% while completion pricing rises 15.6%.
[CORRECTION · WHAT TO DO] Original: “Reforecast monthly spend downward by approximately 2.3% for this model.” Fact: Recalculate prompt and completion costs separately at the captured rates; do not assume a universal decrease.
[CORRECTION · triage.signals] Original: “prompt pricing increased ~19.5%” Fact: Prompt pricing decreases 19.6%.
Stood 121h 25m · cause extraction error · every correction ↗
Captured arithmetic only; vendor rate validity, applicability and effective conditions remain unverified.
WHAT CHANGEDOpenRouter changed pricing for deepseek/deepseek-chat (DeepSeek V3) effective now: prompt tokens from $0.32/M to $0.2574/M (-19.6%), completion tokens from $16,384/M to $16,000/M (-2.3%). Maximum completion tokens reduced from 16,384 to 16,000.
WHY IT MATTERSTotal cost per request decreases 2.3% across all workload mixes. The max_completion_tokens reduction means requests that previously generated up to 16,384 tokens will now be truncated at 16,000, potentially cutting off responses mid-generation.
WHAT TO DOCheck application logs for requests approaching the old 16,384 limit; if any exist, adjust chunking or streaming logic to handle the new 16,000 ceiling. Reforecast monthly spend downward by approximately 2.3% for this model.
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-chat name: DeepSeek: DeepSeek V3 -pricing.prompt: 0.00000032 -pricing.completion: 0.00000089 +pricing.prompt: 0.0000002574 +pricing.completion: 0.0000010287 context_length: 163840 -top_provider.max_completion_tokens: 16384 +top_provider.max_completion_tokens: 16000
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 | Pricing changes and output limit reduction directly affect engineering budgets and token allocation. |
| signals | prompt pricing increased ~19.5% · completion pricing increased ~15.6% · max_completion_tokens decreased from 16384 to 16000 |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |