DeepSeek
Latest releases
DeepSeek
- DeepSeek V4 Flash2026-04-24 · Model releaseOriginal release evidenceDate basis: Official API update dateChecked: 2026-07-15
- DeepSeek V4 Pro2026-04-24 · Model releaseOriginal release evidenceDate basis: Official API update dateChecked: 2026-07-15
Current models
Models, versions, and checkable fields
| Model | Version | Release | OpenRouter | Pricing | Benchmark | Public Signals |
|---|---|---|---|---|---|---|
| DeepSeek V4 Pro DeepSeek's current public V4 Pro MoE model listed in the official catalog and released with open weights. | deepseek-v4-pro | 2026-04-24 | OpenRouter listed | OpenRouter · 2026-07-15: $0.435 / $0.87 input / output, per 1M tokens | No reliable public source | 1 runtime / access record / 1 activity snapshot |
| DeepSeek V4 Flash DeepSeek's current public V4 Flash MoE model listed in the official catalog and released with open weights. | deepseek-v4-flash | 2026-04-24 | OpenRouter listed | Official · 2026-07-15: $0.14 / $0.28 OpenRouter · 2026-07-15: $0.098 / $0.196 input / output, per 1M tokens | No reliable public source | 1 activity snapshot |
| DeepSeek V3.2 DeepSeek's current public V3.2 open-weight backbone release. | DeepSeek-V3.2 | Not disclosed | OpenRouter listed | OpenRouter · 2026-07-15: $0.269 / $0.4 input / output, per 1M tokens | No reliable public source | 1 activity snapshot |
| DeepSeek R1 DeepSeek's current public R1 reasoning model family. | DeepSeek-R1 | Not disclosed | OpenRouter listed | OpenRouter · 2026-07-15: $0.7 / $2.5 input / output, per 1M tokens | No reliable public source | 1 runtime / access record / 1 community discussion / 2 activity snapshots |
| DeepSeek Coder V2 DeepSeek's still-public Coder V2 code-focused MoE model line. | DeepSeek-Coder-V2-Instruct | Not disclosed | Not separately listed on OpenRouter | No reliable public source | No reliable public source | 1 activity snapshot |
| DeepSeek VL2 DeepSeek's current public VL2 multimodal understanding family. | deepseek-vl2 | Not disclosed | Not separately listed on OpenRouter | No reliable public source | No reliable public source | 1 activity snapshot |
| Janus Pro DeepSeek's current public Janus Pro unified multimodal understanding and generation model. | Janus-Pro-7B | Not disclosed | Not separately listed on OpenRouter | No reliable public source | No reliable public source | 1 activity snapshot |
OpenRouter
Availability, pricing, and context
| Model | Version | Status | Pricing | Context | Time |
|---|---|---|---|---|---|
| DeepSeek R1 | DeepSeek-R1 | OpenRouter listed | $0.7 / $2.5 | 163840 | 07/15/2026, 08:11 PM UTC |
| DeepSeek R1 | DeepSeek-R1-0528 | OpenRouter listed | $0.5 / $2.15 | 163840 | 07/15/2026, 08:11 PM UTC |
| DeepSeek V3.2 | DeepSeek-V3.2 | OpenRouter listed | $0.269 / $0.4 | 163840 | 07/15/2026, 08:11 PM UTC |
| DeepSeek V4 Flash | deepseek-v4-flash | OpenRouter listed | $0.098 / $0.196 | 1048576 | 07/15/2026, 08:11 PM UTC |
| DeepSeek V4 Pro | deepseek-v4-pro | OpenRouter listed | $0.435 / $0.87 | 1048576 | 07/15/2026, 08:11 PM UTC |
| DeepSeek Coder V2 | DeepSeek-Coder-V2-Instruct | Not separately listed on OpenRouter | Not applicable | Not disclosed | 07/15/2026, 08:11 PM UTC |
| DeepSeek VL2 | deepseek-vl2 | Not separately listed on OpenRouter | Not applicable | Not disclosed | 07/15/2026, 08:11 PM UTC |
| Janus Pro | Janus-Pro-7B | Not separately listed on OpenRouter | Not applicable | Not disclosed | 07/15/2026, 08:11 PM UTC |
Hot Ranking
Linked ranking rows
| Source | Rank | Model | Value | Time |
|---|---|---|---|---|
| OpenRouter weekly token usage | 3 | DeepSeek: DeepSeek V4 Flash | 5,255,375,087,620 tokens Current snapshot | 07/15/2026, 08:16 PM UTC |
| OpenRouter weekly token usage | 7 | DeepSeek: DeepSeek V4 Pro | 2,568,426,192,565 tokens Current snapshot | 07/15/2026, 08:16 PM UTC |
| OpenRouter weekly token usage | 22 | DeepSeek: DeepSeek V3.2 | 360,196,182,636 tokens Current snapshot | 07/15/2026, 08:16 PM UTC |
Public activity counters
Cumulative Hugging Face model-card counters
| Source | Model | Count | Link |
|---|---|---|---|
| huggingface | DeepSeek R1 | 8,888,471 downloads / 13,471 likes | View source |
| huggingface | DeepSeek V4 Flash | 2,761,359 downloads / 1,755 likes | View source |
| huggingface | DeepSeek V3.2 | 1,457,787 downloads / 1,456 likes | View source |
| huggingface | DeepSeek V4 Pro | 1,431,875 downloads / 5,227 likes | View source |
| huggingface | DeepSeek R1 | 712,731 downloads / 2,457 likes | View source |
| huggingface | Janus Pro | 12,410 downloads / 3,631 likes | View source |
| huggingface | DeepSeek Coder V2 | 5,387 downloads / 693 likes | View source |
| huggingface | DeepSeek VL2 | 4,672 downloads / 387 likes | View source |
Public signals
Public activity and review entry points
| Source | Target | Type | Value | Link |
|---|---|---|---|---|
| huggingface | DeepSeek Coder V2 | Public activity | 5387 downloads / 693 likes | View source |
| huggingface | Janus Pro | Public activity | 12410 downloads / 3631 likes | View source |
| huggingface | DeepSeek R1 | Public activity | 712731 downloads / 2457 likes | View source |
| huggingface | DeepSeek R1 | Public activity | 8888471 downloads / 13471 likes | View source |
| huggingface | DeepSeek V3.2 | Public activity | 1457787 downloads / 1456 likes | View source |
| huggingface | DeepSeek V4 Flash | Public activity | 2761359 downloads / 1755 likes | View source |
| huggingface | DeepSeek V4 Pro | Public activity | 1431875 downloads / 5227 likes | View source |
| huggingface | DeepSeek VL2 | Public activity | 4672 downloads / 387 likes | View source |
| huggingface | DeepSeek V4 Flash | Public signal needing review | DeepSeek-V4-Flash returns garbled output when served with FastLLM | View source |
| huggingface | DeepSeek V4 Pro | Model runtime/access signal | DeepSeek model output sometimes contains DSML tool-call markup | View source |
| DeepSeek | Public snapshot discussion | What is DeepSeek doing for so long? | View source | |
| Unconfirmed upcoming DeepSeek model | Public snapshot discussion | DeepSeek announced they are testing a new model | View source | |
| DeepSeek R1 | Public snapshot discussion | One year since DeepSeek-R1 | View source | |
| huggingface | DeepSeek R1 | Model runtime/access signal | When I run on A800 , it throws error that ValueError: FP8 quantized models is only supported on GPUs with compute capability >= 8.9 (e.g 4090/H100), actual = 8.0 | View source |
Evidence gaps
Company-level display boundary
Public-signal coverage is No safe source