# DeepSeek V3: price, benchmarks, context

- **URL canônica:** https://meirinho.com.br/en/ia/modelos/deepseek-v3/
- **Idioma:** en
- **Atualizado:** 2026-11-17

---
**insufficient data** — 3 of 4 benchmarks

- GPQA Diamond — low-resolution caveat (S index)
- GPQA Diamond (OpenRouter) — low-resolution caveat (S index)
- OTIS Mock AIME — saturated benchmark
- τ-bench Verified (companhia aérea) — low-resolution caveat (S index)

Methodology v2 · updated November 21, 2026 · [Technical tie](https://meirinho.com.br/en/ia/modelos/metodologia/#empate-tecnico) · [Minimum of 4 benchmarks](https://meirinho.com.br/en/ia/modelos/metodologia/#minimo-3-benchmarks) · [Methodology](https://meirinho.com.br/en/ia/modelos/metodologia/)

DeepSeek

# DeepSeek V3

Benchmarks, price, context and versions

GPQA Diamond

Epoch AI

Input US$/1M

Third-party provider

Output US$/1M

Third-party provider

Context

tokens

## Direct answer

**Direct answer:** DeepSeek V3 is a model from DeepSeek. 67.6% on GPQA Diamond, per Epoch AI. On OpenRouter, US$ 0.29 per 1 million input tokens and US$ 1.14 per 1 million output tokens.

## ✓ Good for

- running on your own infrastructure (open weights)

## ✕ Not the best for

- technical reasoning on science topics (111th of 195 on GPQA Diamond, per Epoch AI)
- technical reasoning as the model appears through the API (109th of 134 on GPQA Diamond (OpenRouter), in a statistical tie with one other model, per OpenRouter Benchmarks)

This block follows a rule of this section, using source data: if the model is in the top 25% of the metric, the item goes under "Good for"; if it is in the bottom half, it goes under "Not the best for". The middle band appears in neither. The price uses OpenRouter. Context uses OpenRouter. Scores use Epoch AI or OpenRouter Benchmarks, in the latest version we checked.

## Benchmarks

| GPQA Diamond[? what it is](https://meirinho.com.br/en/ia/modelos/glossario/gpqa-diamond/) | 67.6%**highest score on this metric 96.8%111º / 195 · [Epoch AI](https://epoch.ai/benchmarks/gpqa-diamond), latest version we checked |
| --- | --- |
| OTIS Mock AIME[?](https://meirinho.com.br/en/ia/modelos/glossario/otis-mock-aime/) | 37.8%**highest score on this metric 100%126º / 182 · [Epoch AI](https://epoch.ai/benchmarks/otis-mock-aime-2024-2025), latest version we checked |
| GPQA Diamond (OpenRouter)[?](https://meirinho.com.br/en/ia/modelos/glossario/gpqa-diamond-openrouter/) | 66.7%**highest score on this metric 96.5%109º / 134 · [OpenRouter Benchmarks](https://openrouter.ai/benchmarks/gpqa-diamond), latest version we checked |
| τ-bench Verified (airline)[?](https://meirinho.com.br/en/ia/modelos/glossario/tau-bench/) | 43.8%**highest score on this metric 82%120º / 130 · [OpenRouter Benchmarks](https://openrouter.ai/benchmarks/tau2-bench-airline), latest version we checked |

## Input US$/1M · Output US$/1M

Input US$/1M

US$ 0.29

Output US$/1M

US$ 1.14

## Key specs

| Lab | [DeepSeek](https://meirinho.com.br/en/ia/modelos/lab/deepseek/) |
| --- | --- |
| Origin | China |
| Family | DeepSeek V3 provisional data |
| Weights[?](https://meirinho.com.br/en/ia/modelos/glossario/pesos-abertos/) | open |
| License | no data |
| Also known as | no data |
| Context[?](https://meirinho.com.br/en/ia/modelos/glossario/janela-de-contexto/) | 163,840 tokens |
| Parameters | no data |
| Input | text |
| OpenRouter usage[?](https://meirinho.com.br/en/ia/modelos/glossario/uso-no-openrouter/) | 0.02% OpenRouter |
| Downloads (30 days)[?](https://meirinho.com.br/en/ia/modelos/glossario/downloads-hugging-face/) | no data |
| Use cases provisional data | [Text](https://meirinho.com.br/en/ia/modelos/uso/texto/) [Agents](https://meirinho.com.br/en/ia/modelos/uso/agentes/) |
| Dates in sources | no official dateDecember 26, 2024, published, per Epoch AIMarch 24, 2025, listed on OpenRouter |

Collected November 17, 2026

## Version history

- December 1, 2025 [DeepSeek: DeepSeek V3.2](https://openrouter.ai/deepseek/deepseek-v3.2) OpenRouter, listed on OpenRouter
- December 1, 2025 [DeepSeek-V3.2](https://huggingface.co/deepseek-ai/DeepSeek-V3.2) Hugging Face Hub, Hugging Face repo created
- September 29, 2025 [DeepSeek: DeepSeek V3.2 Exp](https://openrouter.ai/deepseek/deepseek-v3.2-exp) OpenRouter, listed on OpenRouter
- September 29, 2025 [DeepSeek-V3.2-Exp](https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp) Hugging Face Hub, Hugging Face repo created
- August 21, 2025 [DeepSeek: DeepSeek V3.1](https://openrouter.ai/deepseek/deepseek-chat-v3.1) OpenRouter, listed on OpenRouter
- August 21, 2025 [DeepSeek-V3.1](https://huggingface.co/deepseek-ai/DeepSeek-V3.1) Hugging Face Hub, Hugging Face repo created
- March 24, 2025 [DeepSeek: DeepSeek V3 0324](https://openrouter.ai/deepseek/deepseek-chat-v3-0324) OpenRouter, listed on OpenRouter
- March 24, 2025 [DeepSeek-V3-0324](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) Hugging Face Hub, Hugging Face repo created
- December 26, 2024 [DeepSeek: DeepSeek V3](https://openrouter.ai/deepseek/deepseek-chat) OpenRouter, listed on OpenRouter

Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch.

## Sources for this model

- Official model page, [DeepSeek](https://api-docs.deepseek.com/quick_start/pricing/).
- Third-party provider pricing, [OpenRouter](https://openrouter.ai/deepseek/deepseek-chat-v3-0324).
- No official launch announcement.
- Benchmarks: each score carries its own source, in the table above.

## FAQ

On OpenRouter, US$ 0.29 per 1 million input tokens and US$ 1.14 per 1 million output tokens.

DeepSeek V3 is a model from DeepSeek, with open weights, used for text, agents.

163,840 tokens, per OpenRouter.

Yes, it has open weights. License: no data.

67.6% on GPQA Diamond, per Epoch AI.

No data: there is no SWE-bench Verified measurement for it in open sources.

It was made by DeepSeek, a lab from China.

provisional data Questions ordered by what shows up in public Google search suggestions (no volume). Answers use source data only.

## Compare with

### Compare with

- [GPT-5 vs DeepSeek V3](https://meirinho.com.br/en/ia/modelos/comparar/deepseek-v3-vs-gpt-5/)

### Models from DeepSeek

- [DeepSeek V3.1](https://meirinho.com.br/en/ia/modelos/deepseek-chat-v3-1/)
- [DeepSeek V3.2](https://meirinho.com.br/en/ia/modelos/deepseek-v3-2/)
- [DeepSeek V3.2 Exp](https://meirinho.com.br/en/ia/modelos/deepseek-v3-2-exp/)

[See lab](https://meirinho.com.br/en/ia/modelos/lab/deepseek/)

## How to cite

If you use this data, cite the page and the date you accessed it.

```
Adriano Meirinho. "DeepSeek V3". meirinho.com.br. https://meirinho.com.br/en/ia/modelos/deepseek-v3/. Accessed November 21, 2026.
```

Third-party data follows the license of each source. The full table is available as [CSV](https://meirinho.com.br/ia/modelos/dados.csv) and [JSON](https://meirinho.com.br/ia/modelos/dados.json).

[#](#)
