DeepSeek V3: price cut input $0.257 → $0.320 per 1M tokens (+24 %), output $1.03 → $0.890 per 1M tokens (-13 %)
On the listed price of DeepSeek V3 changed: input $0.257 → $0.320 per 1M tokens (+24 %); output $1.03 → $0.890 per 1M tokens (-13 %). A later check on confirmed it. Prices as listed in the United States.
What changed
DeepSeek V3 is cheaper on OpenRouter. Source: price page as read.
What it does to a task
Standard tasks, visible output only, no cache discount; month = 20 tasks an hour around the clock.
| Task | Per task, before | Per task, now | Per month, before | Per month, now |
|---|---|---|---|---|
| L1 · Micro-task (classification or strict JSON extraction) | $0.0006 | $0.0007 | $8.64 | $9.61 |
| L2 · Deep analysis over a 50,000-token document | $0.044 | $0.054 | $642.59 | $781.83 |
| L3 · Agentic coding loop, 15 steps | $0.348 | $0.429 | $5,078.87 | $6,262.96 |
Same task on other models
The L2 task on other models we follow closely, at today's prices:
- Anthropic: Claude Haiku 5.5 — $0.017 per L2 task
- DeepSeek: DeepSeek V4.1 Flash 1017 — $0.051 per L2 task
- Mistral: Mistral Large 4 — $0.118 per L2 task
- Google: Gemini 3.8 Flash — $0.131 per L2 task
Recalculate for your workload
Open the task cost calculator with this model and your own numbers. Model card with the full price history