DeepSeek V3.1 Terminus
DeepSeekReleased Sep 22, 2025Open weights · MITdeepseek-v3.1-terminus
- Reasoning
- yes
- Knowledge cutoff
- March 2026
- Context
- 164K tokens
- Max output
- 64k tokens
- Input
- text
- API features
- tools, structured outputs, JSON mode
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from DeepSeek · cutoff: aiknowledgecutoff.com
- Parameters
- 671B
- 37B active per token (MoE)
- Memory needed
- 403 GB
- in INT4 · 805 GB in INT8
- Smallest cloud setup
- 8× H100
- 640 GB · $17.20/h, about $12,556 a month
- Worth it above
- 10B tokens
- a month, against the cheapest API at $1.27 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 403 GB | 8× H100 · $17.20/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 805 GB | multi-node, over 640 GB | near-lossless |
| FP16 (full) | 1610 GB | multi-node, over 640 GB | reference quality |
At home: 403 GB in INT4 is beyond any desktop or Mac in our machines table, so it means a multi-GPU server.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| Novita AIprecision not stated · 128K ctx (160K) · 32K out | $0.27 | $1.00 | $0.14 | — | — | — | Sep 28 |
| AtlasCloudprecision not read: host terms · 128K ctx (160K) · 64K out | $0.30 | $1.00 | $0.14 | — | — | 99.8% | Oct 5 |
| SiliconFlowfp8 · 160K ctx | $0.27 | $1.00 | — | 18 | 1.90 s | 98.2% | Oct 23 |
| StreamLakeprecision not read: host terms · 125K ctx (160K) · 32K out−40% | $0.34 | $1.03 | — | 45 | 1.45 s | 100% | Oct 23 |
4 of 4 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · −N%: a discount off the host's list price · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| SiliconFlow | fp8 | 160K | not read: host terms | yes | yes | yes |
| Novita AI | not stated by the host | 128K | 32K | — | — | — |
| AtlasCloud | not read: host terms | 128K | 64K | — | — | — |
| StreamLake | not read: host terms | 125K | 32K | no | yes | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
DeepSeek V3.1 Terminus spec sheet, field by field, with sources
DeepSeek V3.1 Terminus spec sheetmaker-declared
What DeepSeek declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 163,840 tokens
- Parameters
- 684,531,386,000≈ 684.5B
- Active experts
- 8 of 256 active per token
- Weight precision
- fp8
- License
- MIT — open weights, permissivecurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ March 2026 third-party estimate — not the maker's figurecurated
- Modalities
- text
- Max output (official)
- to verify
- Release date
- to verify≈ September 22, 2025 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- DeepseekV3ForCausalLM
- Layers
- 61
- Vocabulary
- 129,280 tokens
- Weights first published
- September 22, 2025
- Weights last updated
- September 29, 2025
- Paper
- arXiv:2412.19437
deepseek-ai/DeepSeek-V3.1-Terminus · config.json @ 19510d6 · read 2026-10-08 · 6 of 10 fields published by the maker
| Date | Provider | Input | Output | Change |
|---|---|---|---|---|
| Sep 29 | AtlasCloud | $0.30 → $0.30 | $0.95 → $1.00 | +4% |
Every published change of input or output price, newest first · change on input + output · full series: pricing-history.csv.
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| SiliconFlow | $0.77 | $2.66 | $2.08 |
| Novita AI | $0.77 | $2.66 | $2.08 |
| AtlasCloud | $0.80 | $2.90 | $2.20 |
| StreamLake | $0.86 | $3.25 | $2.40 |
What each host bills for 1,000 requests of a representative job
Your own workload
Other DeepSeek models · all its models, hosts and history ›
- DeepSeek V4.1 Flash29 providers · from $0.47
- DeepSeek V4 Flash 073125 providers · from $0.21
- DeepSeek V4 Pro 081323 providers · from $2.50
- DeepSeek V4.1 Flash 101722 providers · from $0.39
- DeepSeek V4 Pro 042315 providers · from $2.40
- DeepSeek V3.214 providers · from $0.52