---
type: "post"
id: "change-log-deepseek-api-docs-a5a148"
slug: "change-log-deepseek-api-docs-a5a148"
title: "Change Log | DeepSeek API Docs"
description: "DeepSeek says it has upgraded V4.1 Flash to snapshot 1017 across its API, web, and app."
retrieval_nugget: "DeepSeek says it has upgraded V4.1 Flash to snapshot 1017 across its API, web, and app. The update reports higher scores on several benchmarks and shorter task runtime and output, while API access and pricing stay the same; builders should treat those performance claims as a reason to test, not as independent proof of improved production results."
published_at: "2026-11-01"
updated_at: "2026-11-01"
record_date: "2026-11-01"
date_kind: "published_at"
topics: ["claude","claude-code"]
entities: ["DeepSeek","DeepSeek-V4.1-Flash-1017","DeepSeek Harness","Tencent CodeBuddy","WorkBuddy","Claude Code","Mini-SWE-V2"]
related_patterns: ["agent-economics-moves-to-completed-work"]
novelty_label: "same_event"
what_changed: "Same event as DeepSeek обновил V4.1 Flash и открыл API для поиска DeepSeek-V4.1-Flash-1017 уже работает на сайте и в API."
prior_refs: ["memory:fdf3300e-2983-884c-aa26-fa70854d7aa3"]
source_url: "https://api-docs.deepseek.com/updates"
source_urls: ["https://api-docs.deepseek.com/updates"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:change-log-deepseek-api-docs-a5a148"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/change-log-deepseek-api-docs-a5a148/","markdown":"https://newruntime.com/posts/change-log-deepseek-api-docs-a5a148.md","json":"https://newruntime.com/posts/change-log-deepseek-api-docs-a5a148.json"}
---

# Change Log | DeepSeek API Docs

## Retrieval answer

DeepSeek says it has upgraded V4.1 Flash to snapshot 1017 across its API, web, and app. The update reports higher scores on several benchmarks and shorter task runtime and output, while API access and pricing stay the same; builders should treat those performance claims as a reason to test, not as independent proof of improved production results.

## Summary
DeepSeek says it has upgraded V4.1 Flash to snapshot 1017 across its API, web, and app. The update reports higher scores on several benchmarks and shorter task runtime and output, while API access and pricing stay the same; builders should treat those performance claims as a reason to test, not as independent proof of improved production results.

## What happened
DeepSeek’s API documentation change log, dated October 17, says DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017. The company says its official web and app models were updated to the new version as well. API calling and pricing remain unchanged: developers continue to use the model name `deepseek-flash` to access the latest model. The announcement does not describe a new access tier or a migration requirement for API users.

DeepSeek characterizes the update as a performance-cost Pareto improvement over the previous V4.1. Its listed scores include HLE without tools at 42.6, with 44.0 reported for the pure-text subset; Terminal-Bench 4.0 at 42.3; DeepSWE v1.1 at 77.0; ProgramBench at 29.1; NL2Repo-Bench at 68.2; and Automation-Bench v1.0.6 at 60.5. Other reported results include Agents’ Last Exam at 35.4, Apex Bench v1.1 at 72.1, Chartography with tools at 83.0, BabyVision with tools at 92.5, and ZeroBench-main with tools at 51.0. DeepSeek says ZeroBench-main used Pass@5 and marks the alternate HLE score as tested only on the benchmark’s pure-text subset.

The company says it reduced task runtime and output length alongside benchmark improvements, allowing tasks to be completed at lower cost and higher speed. It does not provide a dollar-per-task comparison, a timing table, or a token-usage breakdown in this change-log entry. DeepSeek says the snapshot was specifically optimized for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy, and other platforms. For the best experience with the new model, it recommends DeepSeek Harness 0.2.1-rc.1 or later.

The source identifies the frameworks used for two results: Terminal-Bench 4.0 was tested with Claude Code, and DeepSWE was tested with Mini-SWE-V2. It does not provide fuller run settings in this entry. The announcement is therefore a provider’s release note and benchmark report, not an independently reproduced comparison. For API builders, the immediate operational change is simply that the existing model name now resolves to the upgraded snapshot.

## What the archive adds
[DeepSeek V4.1 Flash 1017 is on AI Gateway.](https://newruntime.com/posts/deepseek-v4-1-flash-1017-is-on-ai-d34087/) covered the snapshot’s availability and cautioned that its task-focused claims needed testing on real workloads. [DeepSeek V4.1 Flash 1017 is now included in Vercel Token Pass.](https://newruntime.com/posts/deepseek-v4-1-flash-1017-is-now-included-323f0e/) and [DeepSeek V4.1 Flash from @deepseek_ai is now available in OpenClaw.](https://newruntime.com/posts/deepseek-v4-1-flash-from-deepseek-ai-is-baea9f/) documented additional availability. This is a new official benchmark and performance claim about a known snapshot, not a first announcement of the model.

## Trend signal
This update weakly supports [agent economics moves to completed work](https://newruntime.com/patterns/agent-economics-moves-to-completed-work/): DeepSeek frames shorter runtime and output as lower cost to complete tasks, rather than only reporting model capability scores. The change log does not quantify cost per verified task or establish that the gains generalize across workloads, so the evidence remains a vendor claim rather than a measured economic result.

## Connections to other events
The archive’s earlier coverage describes this snapshot reaching AI Gateway, Vercel Token Pass, and OpenClaw. The present change-log entry adds DeepSeek’s own performance account to that availability trail: the model can already be reached through those integrations, while the new evidence here is about claimed benchmark and efficiency changes rather than another access expansion.

## Why it matters
For teams using `deepseek-flash`, there is no stated API migration, price change, or access change to plan around. The practical question is whether the snapshot’s reported improvements translate into better outcomes for the team’s own coding and agent tasks. Shorter outputs and runtime could matter even without a new list price, but the source gives no dollar or latency measurements with which to estimate that effect.

The benchmark table is most useful as a set of test candidates, not a deployment guarantee. Teams can compare the listed tasks with their own workload mix, then evaluate quality, retries, latency, and total spend using a fixed harness and representative prompts. The stated use of Claude Code for Terminal-Bench 4.0 and Mini-SWE-V2 for DeepSWE also matters: benchmark scores depend on the surrounding harness, and the announcement does not establish that every runtime will reproduce them. The recommendation to use a newer DeepSeek Harness version is a practical compatibility clue, but not a substitute for local validation.

## Limits and uncertainty
The change log is DeepSeek’s own account; the supplied material includes no independent replication or comparison against other models. It gives scores but not full benchmark configurations, sample sizes, uncertainty ranges, or task-level results. The source does not quantify how much runtime or output length fell, what tasks cost before and after, or whether results vary by harness or prompt. It also does not establish that benchmark improvements will improve a particular team’s production success rate. API price and calling method are explicitly unchanged, so lower cost is presented as an effect of shorter execution and output, not a billing-rate reduction.

## Skill to build now
Create a fixed regression set for the agent tasks that matter to your team and run it through `deepseek-flash` before and after the snapshot change. Record task completion, retry count, wall-clock duration, output tokens, and total spend. Keep the harness, prompts, tools, and acceptance rules stable; otherwise a change in the surrounding system can be mistaken for a model gain.

## Reader takeaway
DeepSeek says snapshot 1017 improves benchmark performance and reduces runtime and output length without changing the `deepseek-flash` API name or price. Existing users can test it without an announced migration, but should measure their own task completion quality and total cost before treating the reported Pareto improvement as a production result.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)