DeepSeek V4.1 Flash 1017 API
The 1017 update of DeepSeek V4.1 Flash for long, input-heavy agentic coding and tool use. It reads images and text natively with a 1M-token context and finishes tasks faster with shorter outputs.
deepseek/deepseek-v4.1-flash-1017
Top up $20 and we add 50%. That gives you about 50.0M input tokens extra on this model, on us.
Create an accountPricing
USD per 1M tokens| Input | $0.20 |
| Output | $0.60 |
| Cached input | $0.0290% below fresh input |
Billed per token with no minimum and no monthly fee. Price your workload
Capabilities and limits
- Context window
- 1M tokens
- Max output
- 375K tokens
- Modality
- multimodal
- Tool calling
- Supported
- Vision
- Supported
- Reasoning
- Optional
- Reasoning effort
- low, high, max (default: high)
- Prompt caching
- Supported (automatic)
Measured performance
7d- Time to first token (p50)
- 12 s
- Time to first token (p95)
- 12 s
- Throughput (avg)
- 29.0 tok/s
- Throughput (p50)
- 50.0 tok/s
Measured on real Run BiOS traffic over the trailing 7d. Your figures will vary with prompt shape and region.
curl https://api.runbios.ai/v1/chat/completions \
-H "Authorization: Bearer $BIOS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash-1017",
"messages": [{ "role": "user", "content": "Hello" }]
}'Existing OpenAI SDK code works by changing the base URL and the key. API overview
Model card
Curated by Run BiOS · benchmark figures, where present, are the card’s own, not our measurements
DeepSeek-V4.1-Flash-1017 is the October 2026 update of DeepSeek-V4.1-Flash, the multimodal Mixture-of-Experts (MoE) model built by DeepSeek (DeepSeek-AI) and described in the technical report "DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression" (2026). It keeps the same 552B-backbone architecture, natively processes images and text and supports contexts of up to one million tokens.
DeepSeek describes 1017 as "a minor update to DeepSeek-V4.1-Flash, delivering substantial improvements across reasoning, coding, agentic tasks and visual benchmarks." Alongside those gains, DeepSeek states that the model "reduces task execution time and output length, enabling users to complete tasks faster and at lower cost and advancing the performance: cost Pareto frontier across multiple agent harnesses."
On the benchmarks DeepSeek publishes for this release, 1017 scores higher than DeepSeek-V4.1-Flash on all eleven of them. The largest gains are on Terminal-Bench 4.0 (+11.1 points), ProgramBench (+8.8), Apex Bench v1.1 (+7.5), HLE without tools (+5.8) and Automation-Bench (+5.7). Against the frontier models in the same table it takes the top score outright on DeepSWE v1.1, Automation-Bench v1.0.6, Agents' Last Exam and Apex Bench v1.1. It does not lead everywhere: the top score goes to Opus 5 on HLE, Terminal-Bench 4.0, ProgramBench, NL2Repo-Bench, Chartography and BabyVision and to GPT-5.6 Sol on ZeroBench.
What changed from DeepSeek-V4.1-Flash
The published model configuration of DeepSeek-V4.1-Flash-1017 is identical to that of DeepSeek-V4.1-Flash: same architecture, same parameter counts, same context length. The change is in the trained weights. DeepSeek has not published separate training details for 1017, so the architecture section below describes both models and the training description belongs to the V4.1-Flash release.
deepseek/deepseek-v4.1-flash remains available on Run BiOS and keeps serving the original DeepSeek-V4.1-Flash weights. Choose this id to use the 1017 update. It is pinned: it will always serve the 1017 weights.
Architecture
DeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organised as a 20-layer causal encoder followed by a 20-layer decoder. The decoder's global KV cache is projected from the final encoder hidden states, which yields an activation profile of 8B parameters per token during prefill and 16B during decode. SWA Bounded Replay reconstructs missing sliding-window-attention KV states by replaying only the most recent tokens of the window, so SWA KV never needs to be persisted to SSD.
Compressed Sparse Attention 2 (CSA2) assigns each attention layer one of three static modes (Full, Reindex or Reuse) to share KV and indexer keys across layers and reuse Top-K sparse-attention indices. A Hierarchical Sparse Indexer bounds deeper indexer cost independently of context length. With FP4 main KV caching, these designs bring the global KV cache to 890 bytes per token, roughly a quarter of DeepSeek-V4-Flash, which DeepSeek positions as substantially improving cost efficiency for input-heavy agentic workloads.
Further published components:
- Single-Pass mHC: revised residual-stream mixing with an efficient Mega-mHC kernel.
- Engram conditional memory: 196B parameters, sparsely accessed via token-based lookup.
- DSpark speculative decoding: semi-autoregressive draft generation with confidence-scheduled verification.
- MoE layers use 1 shared expert and 384 routed experts, activating 6 routed experts per token.
Multimodal path. A vision encoder (DeepSeek-ViT, trained from scratch) and a two-layer MLP projector convert images into visual embeddings that are processed jointly with text from the start of pre-training: images are native to the model, not added afterwards.
Benchmarks
Reproduced from DeepSeek's official DeepSeek-V4.1-Flash-1017 model card. All numbers are as published. Bold marks the highest score in each row.
| Benchmark | DS-V4.1-Flash-1017 | DS-V4.1-Flash | Kimi K3 | GLM-5.3 | GPT-5.6 Sol | Opus 5 |
|---|---|---|---|---|---|---|
| HLE w/o tools | 42.6 (44.0†) | 36.8 (39.1†) | 43.5 | 42.0† | 44.5 | 56.3 |
| Terminal-Bench 4.0 | 42.3 | 31.2 | 12.6 | 37.9 | 39.9 | 51.8 |
| DeepSWE v1.1 | 77.0 | 74.2 | 67.5 | 66.9 | 73.0 | 74.0 |
| ProgramBench | 29.1 | 20.3 | 17.5 | 19.0 | 23.0 | 37.0 |
| NL2Repo-Bench | 68.2 | 65.4‡ | 58.0 | 58.0 | 56.8 | 75.3 |
| Automation-Bench v1.0.6 | 60.5 | 54.8 | 46.7 | 48.8 | 45.8 | 50.3 |
| Agents' Last Exam | 35.4 | 31.8 | 27.6 | 28.5 | 26.7 | 28.6 |
| Apex Bench v1.1 | 72.1 | 64.6 | 47.8 | 56.6 | 54.3 | 65.8 |
| Chartography (w/ tools) | 83.0 | 78.9 | 68.1 | n/a | 79.9 | 84.0 |
| BabyVision (w/ tools) | 92.5 | 89.6 | 85.7 | n/a | 88.9 | 94.1 |
| ZeroBench-main (w/ tools) | 51.0 | 49.0 | 41.0 | n/a | 53.0 | 52.0 |
† Text-only subset of HLE. DeepSeek uses the Claude Code harness for Terminal-Bench 4.0 and the mini-SWE v2 harness for DeepSWE v1.1.
‡ As published in the 1017 model card. DeepSeek's original DeepSeek-V4.1-Flash model card reports 64.0 on NL2Repo-Bench.
For DeepSeek-V4.1-Flash's full published results, including base-model evaluations and performance across agent scaffolds, see the DeepSeek V4.1 Flash model page.
Reasoning Effort and Sampling
DeepSeek exposes a continuously controllable reasoning effort on the V4.1-Flash family that trades inference cost for accuracy. On Run BiOS this model is served through the platform's standard effort ladder (low, high and max), with high applied when a request does not specify one and reasoning can be switched off per request.
DeepSeek's recommended sampling parameters for 1017 are temperature = 1.0 and top_p = 1.0, with a 1M-token context window and max_tokens of at least 256K.
Prompt Encoding
Like DeepSeek-V4.1-Flash, this release ships no Jinja chat template. DeepSeek publishes a self-contained Python reference encoder (covering multi-turn conversations, tool calling, thinking mode, numeric reasoning effort, mid-conversation system messages and interleaved images) and deepseek-recipe, a set of Rust libraries with Python bindings for the same prompt format. Calling this model through the Run BiOS API means none of that applies to you: send an ordinary OpenAI-shaped Chat Completions request and the platform handles the encoding.
Safety Notes
DeepSeek's official model card for DeepSeek-V4.1-Flash-1017 does not publish a dedicated safety evaluation, red-teaming or responsible-use section. The model is released under the MIT License. As with any general-purpose model, outputs should be reviewed before use in high-stakes applications.
On Run BiOS
On the Run BiOS platform DeepSeek-V4.1-Flash-1017 is available as the serverless model deepseek/deepseek-v4.1-flash-1017 with a 1,048,576-token context window, a 384,000-token maximum output, native image input, function calling, structured outputs, optional reasoning at effort low, high or max and automatic prompt caching for repeated prefixes.