DeepSeek has simply launched DeepSeek-V4.1-Flash-1017, a brand new checkpoint of its V4.1-Flash mannequin. It’s live on the DeepSeek website, app and API. DeepSeek claims stronger scores, sooner activity completion and shorter outputs than the September V4.1-Flash release. The corporate calls it a Pareto enchancment in efficiency and value. Alongside it, DeepSeek opened its Search and Fetch APIs to each developer. For agent builders, which means a less expensive coding mannequin plus a first-party net layer behind one API key.
TL;aDR
- Dimension: 552B spine parameters plus 196B Engram parameters. It prompts 8B parameters per token in prefill and 16B in decode. Context is 1M tokens, with as much as 384K output tokens.
- Runs on: DeepSeek API as
deepseek-flash. September V4.1-Flash weights (763B parameters, FP8) are MIT-licensed on Hugging Face. {Hardware} necessities and 1017 weights: Not disclosed. - Efficiency: Beats the earlier V4.1-Flash on all 11 benchmarks DeepSeek reported, by 2.0 to 11.1 factors.
- Greatest: 77.0 on DeepSWE v1.1, above Claude Opus 5 (74.0) in DeepSeek’s desk.
- Worst: 42.6 on HLE with out instruments, versus 56.3 for Claude Opus 5.
What’s DeepSeek-V4.1-Flash-1017?
DeepSeek-V4.1-Flash-1017 is an up to date post-training checkpoint of DeepSeek-V4.1-Flash, served by the deepseek-flash mannequin title. The structure particulars under come from the V4.1-Flash technical report.
DeepSeek group states that DeepSeek-V4.1-Flash-1017 makes use of reinforcement studying with self-feedback for end-to-end undertaking work. Its examples run from sport creation to 3D modeling. It’s tuned for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy. DeepSeek recommends DeepSeek Harness v0.2.1-rc.1 or later.
How does the V4.1-Flash structure work?
V4.1-Flash is a multimodal Mixture-of-Experts model with 40 Transformer layers. They cut up right into a 20-layer causal encoder and a 20-layer decoder.
- Causal Encoder-Decoder (CED): Decoder international KV is projected from the ultimate encoder hidden states. This almost halves prefill compute.
- Compressed Sparse Consideration 2 (CSA2): Layers run in Full, Reindex or Reuse mode. They share most important KV and indexer keys throughout layers.
- FP4 most important KV cache: World KV drops to 890 bytes per token, about 1/4 of V4-Flash.
- SWA Bounded Replay: Persistent KV cache shrinks to roughly 1/8 of V4-Flash.
- Reasoning effort: A 1 to 100 scalar. The API tiers low, excessive and max map to 50, 75 and 100.
DeepSeek-V4.1-Flash-1017 benchmarks
All scores under are DeepSeek’s own figures. DeepSeek ran Terminal-Bench 4.0 in Claude Code and DeepSWE in Mini-SWE-V2. ZeroBench-main makes use of Go@5.
| Benchmark | V4.1-Flash-1017 | V4.1-Flash | Kimi K3 | GLM 5.3 | GPT 5.6-Sol | Claude Opus 5 |
|---|---|---|---|---|---|---|
| DeepSWE v1.1 | 77.0 | 74.2 | 67.5 | 66.9 | 73.0 | 74.0 |
| Terminal-Bench 4.0 | 42.3 | 31.2 | 12.6 | 37.9 | 39.9 | 51.8 |
| Automation-Bench v1.0.6 | 60.5 | 54.8 | 46.7 | 48.8 | 45.8 | 50.3 |
| Brokers’ Final Examination | 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 |
| 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 |
| HLE (no instruments) | 42.6 / 44.0* | 36.8 / 39.1* | 43.5 | 42.0* | 44.5 | 56.3 |
| BabyVision (w/ instruments) | 92.5 | 89.6 | 85.7 | Not disclosed | 88.9 | 94.1 |
*Textual content-only HLE subset.
The 1017 checkpoint leads on DeepSWE, Automation-Bench, Brokers’ Final Examination and Apex Bench. Claude Opus 5 nonetheless leads on Terminal-Bench 4.0, ProgramBench, NL2Repo and HLE.
window.addEventListener(‘message’,operate(e){if(e.information&&e.information.dsxHeight){var f=doc.getElementById(‘dsx-1017-frame’);if(f&&e.supply===f.contentWindow){f.type.top=e.information.dsxHeight+’px’;}}});
<!– Fallback: add dsv41-1017-explainer.html to /wp-content/uploads/ through FTP and use: –>
How does it evaluate with open-weight rivals?
| Characteristic | DeepSeek-V4.1-Flash-1017 | Kimi K3 | GLM-5.3 |
|---|---|---|---|
| Complete parameters | 552B spine + 196B Engram | 2.8T | 753B |
| Energetic parameters | 8B prefill / 16B decode | 104B | Not disclosed |
| Context | 1M | 1,048,576 | Not disclosed |
| License | MIT (V4.1-Flash); 1017 weights Not disclosed | Kimi K3 License | GLM-5.3 license |
| Modality | Textual content, picture in; textual content out | Textual content, picture | Textual content |
| DeepSWE v1.1 | 77.0 | 67.5 | 66.9 |
| Brokers’ Final Examination | 35.4 | 28.3 | 28.5 |
| API worth (1M enter miss / output, peak) | $0.30 / $1.20 | Not disclosed | Not disclosed |
What are DeepSeek’s Search and Fetch APIs?
The Search API returns ranked net outcomes for a question. The Fetch API returns a single web page as Markdown. DeepSeek says its in-house engine serves tons of of tens of millions of requests each day.
- Endpoints:
POST /v1/instruments/searchandPOST /v1/instruments/fetchonapi.deepseek.com. - Search choices: 1 to 50 outcomes, non-obligatory web page content material, and operators like
website:,lang:,after:andearlier than:. - End result fields: Every consequence has a 0 to three rerank rating and a 0 to 4 authority tier.
- Worth: $0.75 per 1,000 successful requests for every endpoint. Empty outcomes and failures are free.
- Charge restrict: 5 QPS per account for every API.
- Free window: Server-side calls through DeepSeek Harness, Claude Code or Codex keep free till November 1, 2026.
Deployment and pricing
Name the mannequin with deepseek-flash by OpenAI or Anthropic-format endpoints. Peak pricing is $0.30 per 1M cache-miss enter tokens and $1.20 per 1M output tokens. Off-peak charges are half.
Key Takeaways
- 1017 scores 77.0 on DeepSWE v1.1, up from 74.2.
- It leads Automation-Bench (60.5) and Brokers’ Final Examination (35.4).
- Claude Opus 5 nonetheless wins on HLE and Terminal-Bench 4.0.
- Search and Fetch value $0.75 per 1,000 calls every.
- Agent harness calls keep free till November 1, 2026.
Try the DeepSeek announcement, the Search and Fetch docs, the technical report and the model card. All credit score goes to the researcher of this undertaking. Additionally, be happy to comply with us on Twitter and don’t overlook to affix our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
The put up DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API appeared first on MarkTechPost.

