🔍 Read the full analysis: Is DeepSeek V4.1 Flash 1017 The Right Value For Your AI Agents? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
DeepSeek released V4.1 Flash 1017 on October 17, replacing the September model on its website, app and API without changing the listed prices. Its company-reported benchmarks show gains, especially on some multi-step agent tasks, but the results have not been independently reproduced and the comparison chart omits newer rivals.
The update is served under the API name deepseek-flash; DeepSeek says older names, including deepseek-v4-flash and deepseek-v4-flash-vision-exp, also route to the new version. The model’s weights are available under the MIT license. DeepSeek describes the release as a minor update and has not said precisely what changed in training.
In DeepSeek’s published table, Terminal-Bench 4.0 rises from 31.2 to 42.3, while ProgramBench increases from 20.3 to 29.1. DeepSWE v1.1, a coding benchmark, moves from 74.2 to 77.0. These are the company’s figures, not independently verified results. DeepSeek says the newer model produces shorter outputs, which it says can make tasks faster and cheaper.
The model card lists API rates per million tokens of $0.15 for uncached input and $0.60 for output off-peak, rising to $0.30 and $1.20 during peak hours. Cached input is priced at $0.003 off-peak and $0.006 at peak. DeepSeek defines weekday peak periods as 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays; it says batch jobs outside those periods cost half the listed rate.
Infographic · DeepSeek V4.1 Flash 1017
Better on every benchmark, a fraction of the cost per task, and still behind the frontier.
DeepSeek’s October update improves all eleven scores it published over the September version. Against the four rivals DeepSeek picked, it leads four benchmarks and trails seven. Its real advantage is the price of a finished task.
1017 vs the September version
Score per benchmark (0–100), sorted by gain. Hover a row for the exact values.
- Terminal-Bench 4.042.331.2+11.1
- ProgramBench29.120.3+8.8
- Apex Bench v1.172.164.6+7.5
- HLE (no tools)42.636.8+5.8
- Automation-Bench60.554.8+5.7
- Chartography83.078.9+4.1
- Agents' Last Exam35.431.8+3.6
- BabyVision92.589.6+2.9
- DeepSWE v1.177.074.2+2.8
- NL2Repo-Bench68.265.4+2.8
- ZeroBench-main51.049.0+2.0
Against the best rival per benchmark
Points ahead of or behind the strongest of Kimi K3, GLM-5.3, GPT-5.6 Sol and Claude Opus 5 on each row.
- Automation-Benchbest rival: Claude Opus 5 50.3+10.2
- Agents' Last Exambest rival: Claude Opus 5 28.6+6.8
- Apex Bench v1.1best rival: Claude Opus 5 65.8+6.3
- DeepSWE v1.1best rival: Claude Opus 5 74.0+3.0
- Chartographybest rival: Claude Opus 5 84.0-1.0
- BabyVisionbest rival: Claude Opus 5 94.1-1.6
- ZeroBench-mainbest rival: GPT-5.6 Sol 53.0-2.0
- NL2Repo-Benchbest rival: Claude Opus 5 75.3-7.1
- ProgramBenchbest rival: Claude Opus 5 37.0-7.9
- Terminal-Bench 4.0best rival: Claude Opus 5 51.8-9.5
- HLE (no tools)best rival: Claude Opus 5 56.3-13.7
Cost per DeepSWE task
USD per task, linear scale to $10, as read from DeepSeek’s chart.
- 1017, low effortabove 72% of DeepSWE solvedabout $0.10
- 1017, default77.0% of DeepSWE solvedjust under $0.20
- V4.1 Flash (September)74.2% of DeepSWE solvedabout $0.50
- Frontier models on the chartGPT-6 Astra, Gemini 3.8 Flash, Claude Opus 5$2 to $10+
None of the frontier models on DeepSeek’s chart reached 77%.
API price per 1M tokens
| Token type | Off-peak | Peak |
|---|---|---|
| Input, cache miss | $0.15 | $0.30 |
| Input, cache hit | $0.003 | $0.006 |
| Output | $0.60 | $1.20 |
Peak: weekdays 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays.
All scores are DeepSeek’s own and not independently reproduced; rivals are the ones DeepSeek chose (GPT-5.6 Sol and Claude Opus 5, not the current GPT-6 and Opus 5.5 models). Sources: DeepSeek model card, DeepSeek API pricing, cost per task via The Decoder (17–18 Oct 2026).
Powered by Thorsten Meyer AI
Lower Costs for Agent Tasks
For teams running coding and other multi-step agents, the practical question is not just which model tops a benchmark, but how much it costs to complete a usable task. DeepSeek’s reported pricing and its own cost-per-task chart make 1017 a candidate for budget-sensitive workloads, especially if its performance holds in a customer’s environment.
The company’s chart, as reported by The Decoder, puts 1017 at 77% on DeepSWE for just under $0.20 per task; a lower-effort setting scores above 72% at about $0.10. DeepSeek’s September model is shown at 74.2% for about $0.50. These figures are tied to the chart’s setup and should not be treated as a universal cost forecast: real bills depend on prompt length, output, retries, caching and the agent harness.
DeepSeek’s table also does not show 1017 as the strongest model across the board. It leads four of 11 rows, while Claude Opus 5 leads six and GPT-5.6 Sol leads one. That distinction matters for buyers weighing price against task quality: the update may be attractive for workloads where its cost advantage outweighs differences in reasoning, coding or vision performance.
Kingy AI reported an early pilot on seven small tasks, run twice. In that test, 1017, GPT-6.1 Sol and Claude Sonnet 5.5 each passed all 14 runs, with 1017 costing roughly a quarter as much as the other two, according to Kingy AI. The small test cannot establish a broad ranking, and it also found 1017 was slower on one task. It is an early data point, not a substitute for independent, larger evaluations.
As an affiliate, we earn on qualifying purchases.
What DeepSeek Released
DeepSeek-V4.1-Flash-1017 follows the V4.1 Flash release of September 10, 2026. The new version became the default on DeepSeek’s products on October 17, so API users calling the general model alias receive 1017 without changing that alias. The company recommends its own DeepSeek Harness version 0.2.1-rc.1 or later and says it tuned the model for that harness and Tencent’s CodeBuddy and WorkBuddy.
DeepSeek’s technical materials describe a sparse Mixture of Experts model with image input and a one-million-token context window. The report, as summarised by Kingy AI, describes a Causal Encoder-Decoder design with 40 layers and 384 routed experts, six active per token, plus a shared expert. Published size descriptions differ by counting method: the technical report describes a 552-billion-parameter backbone, while Hugging Face lists 763 billion. DeepSeek reports about eight billion active parameters during prefill and 16 billion during decoding.
The technical report is titled Pushing the Limits of KV Cache Compression and reports a cache size of about 890 bytes per token. DeepSeek has not disclosed the exact training changes behind the update. The Decoder reports that the gains may involve post-training reinforcement learning with self-feedback, but that account is not a detailed training disclosure from DeepSeek.
AI model performance testing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evidence Still Has Gaps
Independent validation is limited. The 11 benchmark results in the model card come from DeepSeek, and the source material does not provide independent reproductions of the full table. Some comparisons also use different evaluation setups: DeepSeek says Terminal-Bench ran in the Claude Code harness and DeepSWE in Mini-SWE-V2. Scores should be compared with care when harnesses or test conditions differ.
The rival set is another limitation. DeepSeek’s chart includes GPT-5.6 Sol and Claude Opus 5, but not newer models named in the source material, including GPT-6.1 Sol, GPT-6 Astra, Opus 5.5 and Gemini 4 Argon. Kingy AI cites separate Terminal-Bench results of 57 to 66 for current frontier models, while noting those evaluations are not directly matched to DeepSeek’s run. The figures offer context, not a controlled head-to-head.
It is also unclear what training changes produced the improvements, how consistently they transfer beyond DeepSeek’s recommended agent setups, and whether the reported cost per task will match production use. The supplied source cuts off during its description of a separate DeepSeek search-tools announcement, so its features and availability cannot be established here.
As an affiliate, we earn on qualifying purchases.
Test Results Against Workloads
For developers considering the update, the next useful step is to test 1017 against their own tasks, prompts, tools and failure criteria rather than rely on a single benchmark score. Comparisons should record completion rate, retries, latency and total token cost, and use the same harness and task set across models. The October 17 release is already the default for DeepSeek’s named API alias, and the company recommends DeepSeek Harness 0.2.1-rc.1 or later.
Further independent evaluations could clarify whether the gains on Terminal-Bench and ProgramBench hold outside DeepSeek’s published tests, and whether its cost advantage persists when tasks require longer outputs or repeated attempts. No date for such evaluations, a fuller training disclosure or a subsequent model release is provided in the source material.
As an affiliate, we earn on qualifying purchases.
Key Questions
When did DeepSeek release V4.1 Flash 1017?
DeepSeek released V4.1 Flash 1017 on October 17, 2026. It replaced the September version on the company’s website, app and API.
Did DeepSeek change the API price?
No price change is reported. DeepSeek kept its listed rates, which vary by input type and time of day. Off-peak rates include $0.15 per million uncached input tokens and $0.60 per million output tokens.
Are the benchmark improvements independently verified?
Not in the source material. The published benchmark scores are DeepSeek’s own results; independent reproduction of the full table has not been reported.
Does 1017 lead every benchmark in DeepSeek’s chart?
No. It leads four of 11 benchmark rows. Claude Opus 5 leads six and GPT-5.6 Sol leads one, according to DeepSeek’s table.
What should developers check before using it?
Run representative tasks under the intended agent harness and compare task success, latency, retries and full API cost. Benchmark scores and per-task costs can vary with prompts, tools and evaluation setup.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
