What Does A Tenth Of The Price Buy With DeepSeek V4.1 Flash 1017?

🔍 Read the full analysis: What Does A Tenth Of The Price Buy With DeepSeek V4.1 Flash 1017? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, replacing the September version across its website, app and API while keeping prices unchanged. The company says the update improves results on all 11 benchmarks it published, with the biggest gains on some long-running agent tasks; the figures have not been independently reproduced.

The update is served through the deepseek-flash API name, and older names including deepseek-v4-flash now route to it. DeepSeek describes 1017 as a minor update. Its model card reports Terminal-Bench 4.0 rising from 31.2 to 42.3 and ProgramBench from 20.3 to 29.1. DeepSWE, another coding benchmark, increased more modestly, from 74.2 to 77.0.

DeepSeek also says the newer model produces shorter outputs, which can reduce completion time and token charges. Its published API rates remain $0.15 per million uncached input tokens off-peak and $0.30 at peak, with output priced at $0.60 and $1.20 per million tokens, respectively. Cached input is listed at $0.003 off-peak and $0.006 at peak. DeepSeek’s peak periods are 01:00–04:00 and 06:00–10:00 UTC on weekdays, excluding Chinese public holidays.

The company’s model card compares 1017 with its September model and five selected competitors. In that table, 1017 leads four of 11 rows; Claude Opus 5 leads six and GPT-5.6 Sol leads one. DeepSeek’s HLE result uses the full test set, while one competitor’s listed figure is text-only, a difference that complicates direct comparison.

Lower Prices, Stronger Agent Runs

The update’s clearest selling point is the potential cost of completing a task, rather than a claim to lead every benchmark. The Decoder reported that DeepSeek’s DeepSWE chart puts 1017 at 77% for just under $0.20 per task; a lower-effort setting scored above 72% at about $0.10. On the same chart, the September model cost about $0.50 per task for a 74.2% score, while the frontier systems shown cost between $2 and more than $10. These are figures from DeepSeek’s evaluation, not a general guarantee of costs across workloads.

For developers running coding agents at scale, lower per-task costs could make repeated testing and longer automated workflows more affordable. The comparison depends on the task, harness, amount of input and output, and success criteria. A model that is cheaper per attempt may not be cheaper overall if it needs more retries or human correction. DeepSeek’s reported shorter outputs could help, but the available figures do not settle that question across ordinary production use.

A small outside test offers limited support for the value proposition. Kingy AI reported that in a pilot involving seven tasks run twice, 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. The pilot also found that 1017 was not always the fastest: one task took about 27 seconds, compared with about three seconds for Sonnet. With all three models scoring perfectly in this small sample, it cannot establish which performs best overall.

What Changed in Version 1017

DeepSeek released the earlier V4.1 Flash model on September 10, 2026. The October 17 release takes its place automatically for users of the main API alias, and the company says the price has not changed. The model’s weights are available under the MIT license on Hugging Face and ModelScope, giving developers an option to run the released checkpoint themselves, subject to their hardware and deployment needs.

According to the technical report, as summarized by Kingy AI, the model uses a sparse mixture-of-experts design with image input and a one-million-token context window. The report describes a 40-layer causal encoder-decoder, with 384 routed experts and six active per token, plus a shared expert. DeepSeek’s reported architecture figures include a 552-billion-parameter backbone, while Hugging Face lists 763 billion using a different counting method. The hosted API supports up to 384,000 output tokens.

DeepSeek has not described what changed in training for 1017. The Decoder reported that the gains may come from post-training with reinforcement learning and self-feedback, but that account is not a detailed explanation from DeepSeek. The company recommends its own DeepSeek Harness version 0.2.1-rc.1 or later and says it tuned the model for Tencent’s CodeBuddy and WorkBuddy as well.

““minor update””

— DeepSeek

Benchmark Comparisons Have Limits

Independent reproduction is lacking for the 11 benchmark results in DeepSeek’s release materials. The company’s table also compares 1017 with rivals it selected, including systems that are not the newest models named in the source material. Separate Terminal-Bench results for newer frontier models were reported between 57 and 66 by Kingy AI, but those evaluations used different setups and are not directly comparable.

It is also unclear which training changes produced the gains: DeepSeek has not disclosed a specific account, and the reported self-feedback explanation comes from The Decoder. The pilot from Kingy AI covered only seven tasks, run twice, so it cannot show how the model performs across varied software projects or larger workloads. No broad, independent cost or speed study is available in the material provided.

Benchmark rankings also depend on harnesses and test conditions. DeepSeek says Terminal-Bench used the Claude Code harness and DeepSWE used Mini-SWE-V2. The HLE figures use different test subsets for some models. Those details limit what can be inferred from a single score or row in the company’s chart.

Independent Tests Will Matter

The next useful evidence will be independent evaluations that compare 1017 with current alternatives on the same tasks, harnesses and scoring rules. Those tests can clarify whether the gains hold outside DeepSeek’s published setup and whether shorter outputs translate into lower total costs after retries and review.

Developers can use the updated hosted model immediately through the existing API alias, while the MIT-licensed weights provide another route for evaluation. Results from real deployments, including task success, latency and operating costs, will help show where the update is useful. DeepSeek has not announced a date for a further model update or a more detailed account of 1017’s training changes.

Key Questions

What is DeepSeek V4.1 Flash 1017?

It is DeepSeek’s October 17, 2026 update to its V4.1 Flash model. It replaces the September version on the company’s website, app and API.

Did DeepSeek raise its API prices?

No. DeepSeek says prices are unchanged. Listed rates vary by input type and time of day; for example, uncached input costs $0.15 per million tokens off-peak and $0.30 at peak.

Does 1017 beat all the competitors in DeepSeek’s tests?

No. In DeepSeek’s own table, 1017 leads four of 11 benchmark rows. Claude Opus 5 leads six and GPT-5.6 Sol leads one. The results have not been independently reproduced.

How much cheaper is 1017 per coding task?

DeepSeek’s DeepSWE chart, as reported by The Decoder, puts 1017 at just under $0.20 per task for a 77% score. The September model’s listed result was about $0.50 per task for 74.2%. Those figures are tied to the reported test and may not predict costs for other workloads.

What remains unverified?

The published benchmark gains and much of the cost comparison lack broad independent testing. The precise training changes behind the update are also not disclosed by DeepSeek.

Source: ThorstenMeyerAI.com