DeepSeek announced DeepSeek-V4.1-Flash-1017, a checkpoint update to V4.1-Flash, and says it is live on the company’s website, app, and API. ChosunBiz reported that DeepSeek unveiled the update on October 17, 2026. The release also made the Search and Fetch APIs publicly accessible, with separate per-request charges.
What changed in the release
DeepSeek describes DeepSeek-V4.1-Flash-1017 as a minor update to V4.1-Flash, with faster task completion and shorter outputs. Its API uses the model name deepseek-flash; DeepSeek lists a 1-million-token context window and a maximum output of 384,000 tokens. Thinking mode is the default, while non-thinking mode and vision are also supported.
DeepSeek’s published benchmark scores
DeepSeek’s model card reports higher scores for the new checkpoint on the three benchmarks shown for both versions here. The company names Claude Code as the Terminal-Bench 4.0 harness, a software setup used to run the evaluation.
| Benchmark | V4.1-Flash | V4.1-Flash-1017 |
| Terminal-Bench 4.0 | 31.2 | 42.3 |
| ProgramBench | 20.3 | 29.1 |
| HLE without tools | 36.8 | 42.6 |
DeepSeek also reports a DeepSWE v1.1 score of 77.0 for V4.1-Flash-1017, compared with 74.2 for V4.1-Flash; the model card names Mini-SWE-V2 as that evaluation’s harness. These are the company’s reported results. On October 18, 2026, ChosunBiz reported that external verification had not yet taken place as of that date.
API charges for tokens, Search, and Fetch
DeepSeek lists token rates in U.S. dollars per 1 million tokens. Input charges vary depending on whether input is cached, and all three rates differ between peak and off-peak periods.
| API charge | Off-peak per 1M tokens | Peak per 1M tokens |
| Cached input | $0.003 | $0.006 |
| Uncached input | $0.15 | $0.30 |
| Output | $0.60 | $1.20 |
Peak hours are weekdays from 01:00 to 04:00 UTC and 06:00 to 10:00 UTC, excluding Chinese public holidays. All other hours, including weekends and those holidays, are off-peak. DeepSeek’s API pricing page lists the current rates and billing terms. NeoTeo’s earlier explanation of DeepSeek V4.1 Flash API pricing covers how cache reuse affects token costs; Search and Fetch add a separate meter.
Each endpoint costs $0.75 per 1,000 successful requests, billed separately from model tokens. As of October 20, DeepSeek says server-side Search and Fetch calls through DeepSeek Harness, Claude Code, and Codex are free. Per-request billing for those calls is scheduled to start at 00:00 Beijing time on November 1, 2026.
Weights and local inference
DeepSeek links the checkpoint weights through Hugging Face and ModelScope. Its model card states that the repository and weights use the MIT License. The linked inference instructions show how to convert the weights and run inference locally. The example sets tensor parallelism to MP=8 and uses fp4 expert weights; it is an example configuration, not a stated minimum hardware requirement.
A creator’s 3D demonstration
AI with Eric’s video shows a robotic-arm camera-mount modification and a voxel-art comparison between the original V4.1 Flash and the new checkpoint. The creator says Qwen3.8-Flash-Next had struggled with the camera-mount task and reports success with DeepSeek-V4.1-Flash-1017. It is a qualitative demonstration of one task, not a standardized benchmark.
