DeepSeek V3.1: token counter & pricing
DeepSeek · approximate, within ±3% of reference · pricing as of 2026-05-31.
- Provider
- DeepSeek
- API model ID
deepseek-ai/DeepSeek-V3.1- Context window
- 128,000 tokens
- Input price
- $0.60 per 1M tokens
- Output price
- $1.70 per 1M tokens
- Tokenizer accuracy
- approximate, within ±3% of reference
- Pricing as of
- 2026-05-31
Open the counter to count tokens for DeepSeek V3.1 in real time.
What is DeepSeek V3.1?
DeepSeek V3.1 is the incremental update to DeepSeek V3, same architecture and tokenizer, improved instruction-following and tool-use. $0.60 input / $1.70 output per 1M tokens via Together.ai.
DeepSeek's direct API offers V3 at $0.27/$1.10 (cheaper). V3.1 isn't currently advertised on the direct DeepSeek API. Together is the listed reseller for V3.1 as of April 2026.
How tokens are counted here
DeepSeek uses a BPE tokenizer derived from the LLaMA family with extensions. We approximate in your browser, accurate within roughly ±3% for typical English text. Marked ≈±3%.
For exact counts: AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V3.1").
Pricing notes
$0.60 input / $1.70 output per 1M (Together.ai).
For 1,000 input + 200 output: $0.00094 per call, $940 per 1M calls.
128K context window.
When to use DeepSeek V3.1
- Cost-sensitive frontier-tier workloads, comparable quality to GPT-5.2 ($1.75/$14) at a third the input cost and 88% lower output cost.
- Open-weights workloads where compliance / fine-tuning / self-hosting matter eventually but you want API access for development.
- Coding workloads. V3.1 benchmarks well on code generation tasks, often beating Llama 3.3 70B.
When not to use it:
- Production with cross-border data flow concerns about China-affiliated providers (Together is US-hosted; the model weights are from DeepSeek in China).
- Workloads requiring the strongest reliability on function-calling and structured outputs, OpenAI / Anthropic still lead.
- Real-time chat with low-latency requirements.
Common questions
DeepSeek V3.1 vs V3?
V3 (direct API): $0.27/$1.10. V3.1 (Together): $0.60/$1.70. The catalog keeps both because they reflect different real-world access paths. If you can use DeepSeek's direct API, V3 is dramatically cheaper. If you need a US-hosted reseller for compliance, V3.1 via Together is the realistic option.
DeepSeek V3.1 vs GPT-5.2?
GPT-5.2 ($1.75/$14) is closer-priced than V3 but still 3× more expensive on input. For tasks where you've measured DeepSeek matching GPT-5.2 quality, the cost gap is real money at scale.
Is V3.1 the same as R1?
No. R1 is the reasoning-tier model (extended chain-of-thought, higher cost, longer latency). V3.1 is the general-purpose conversational/code model. Different positioning, different price.
Compare DeepSeek V3.1 to other models
- DeepSeek V3 (DeepSeek, $0.27/$1.10)
- DeepSeek R1 (DeepSeek, $3.00/$7.00)
- Llama 3.1 70B (Meta, $0.59/$0.79)
- Gemini 3 Flash (Google, $0.50/$3.00)
- GPT-5.4 Mini (OpenAI, $0.75/$4.50)