AI Agents On A Budget: Evaluating DeepSeek V4.1 Flash 1017

🔍 Read the full analysis: AI Agents On A Budget: Evaluating 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 V4.1 Flash 1017 on October 17, replacing the September version across its website, app and API while keeping token prices unchanged. The company reports gains on all 11 benchmarks it published, including larger improvements on some agent tasks, but the results are vendor-reported and not independently reproduced.

The updated model is served under the API name deepseek-flash; DeepSeek says older names, including deepseek-v4-flash, now route to it. The company describes the release as a minor update. Its published table shows Terminal-Bench 4.0 rising from 31.2 to 42.3 and ProgramBench from 20.3 to 29.1. DeepSWE v1.1 moved from 74.2 to 77.0. These are benchmark score changes, not independently verified estimates of performance in everyday use.

DeepSeek says 1017 produces shorter outputs, which can reduce task time and token costs. The model remains available at the same listed rates: input costs range from $0.003 to $0.30 per million tokens, depending on cache status and time of day, while output costs $0.60 off-peak or $1.20 at peak. DeepSeek defines peak weekday periods as 01:00–04:00 and 06:00–10:00 UTC, except Chinese public holidays.

The weights are published under the MIT license, and the model supports image input, according to the source material. DeepSeek’s technical report describes a one-million-token context window and a sparse Mixture of Experts architecture. The Decoder attributes the benchmark gains to post-training with reinforcement learning and model self-feedback, but DeepSeek has not said what changed in training for this release.

The Cost Case for Agent Work

The release’s main practical pitch is cost per completed task, not 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, while a lower-effort setting scored above 72% at about $0.10. Those figures are based on DeepSeek’s chart. They suggest the model could appeal to developers running many coding or automation tasks, where repeated inference costs can add up.

A Kingy AI pilot offers a limited outside check: across seven small tasks run twice, 1017, GPT-6.1 Sol and Claude Sonnet 5.5 each passed all 14 runs, with 1017 reported at roughly a quarter of the other models’ cost. The sample is too small to establish general reliability, and equal scores did not distinguish the models. The pilot also found 1017 was slower on one task, taking about 27 seconds versus about three seconds for Sonnet. For buyers, price, latency and task success all matter; a low token bill alone does not establish the cheapest or best workflow overall.

How the Benchmark Compares

DeepSeek compared 1017 with its September model and four selected rivals: Kimi K3, GLM-5.3, GPT-5.6 Sol and Claude Opus 5. In the company’s table, 1017 led four of 11 rows; Claude Opus 5 led six and GPT-5.6 Sol led one. Opus 5 scored 56.3 on HLE, compared with 42.6 for 1017, and led on ProgramBench and NL2Repo-Bench. The table therefore presents a mixed picture: the update improves on its predecessor, while other models lead several listed tasks.

The comparison also does not include newer models named in the source material, such as GPT-6.1 Sol, GPT-6 Astra, Opus 5.5 or Gemini 4 Argon. Kingy AI cited Terminal-Bench results of 57 to 66 for current frontier models, but those evaluations were conducted separately and are not directly comparable to DeepSeek’s table. DeepSeek’s HLE scores also use the full test set, while GLM-5.3’s listed score is text-only; the company gives 44.0 for 1017 on the text-only subset.

“minor update”

— DeepSeek, in its release description as reported by ThorstenMeyerAI.com

Independent Results Still Needed

No independent reproduction of DeepSeek’s 11 benchmark results is provided in the source material. The figures also depend on evaluation setups: Terminal-Bench was run in the Claude Code harness, while DeepSWE used Mini-SWE-V2. Differences in harnesses, prompts, model versions and scoring can affect comparisons, so results from separate tests should not be treated as a single leaderboard.

It is also unclear what specifically changed in 1017’s training. The account attributes the improvement to reinforcement learning with self-feedback, but this is a report about the likely method, not a detailed change log confirmed by DeepSeek. The independent pilot involved only seven small tasks, and its perfect scores across three models cannot show how they perform across a broad range of real-world work. Actual costs will also vary with input length, cached tokens, output size and whether requests fall in peak hours.

Watching Real-World Agent Use

Developers can test the new version through DeepSeek’s API, app and website; calls to deepseek-flash now reach 1017, according to the report. DeepSeek recommends its own Harness version 0.2.1-rc.1 or later and says the model is tuned for that environment, as well as Tencent’s CodeBuddy and WorkBuddy. More useful comparisons will come from independent tests that publish task sets, costs, runtimes and evaluation settings.

The report also says DeepSeek opened its search engine to developers through two agent tools, Search and Fetch, but the source material ends before describing their functions or availability in detail. Those specifics remain unconfirmed here. For now, the clearest next measure is whether outside evaluations and deployed workflows reproduce the reported quality and cost gains without unacceptable delays or reliability trade-offs.

Key Questions

When did DeepSeek release V4.1 Flash 1017?

DeepSeek released it on October 17, 2026, replacing the September version on its website, app and API.

Did DeepSeek change the model’s prices?

No price change was reported. The listed rates vary by input cache status and time of day; output is listed at $0.60 per million tokens off-peak and $1.20 at peak.

Are the reported benchmark gains independently verified?

Not in the source material. The 11 benchmark results are reported by DeepSeek. A small Kingy AI pilot tested seven tasks, but it is not enough to verify broad performance.

Does 1017 lead every benchmark in DeepSeek’s table?

No. It led four of the 11 rows. Claude Opus 5 led six, and GPT-5.6 Sol led one, according to DeepSeek’s published comparison.

What should developers check before adopting it?

Test it on their own tasks and compare completion quality, latency and total cost. Costs depend on token use, cache hits and peak-hour pricing, while the published benchmark results have not been independently reproduced.

Source: ThorstenMeyerAI.com