DeepSeek's lineup has an odd shape this autumn. V4 Pro 0813 is the bigger, pricier model: at peak hours its output costs $3.96 per 1M tokens, 3.3 times the Flash's $1.20. But the Flash has been retrained twice since Pro's August release, and the October 17 update, V4.1 Flash 1017, now outscores Pro on every agentic and coding benchmark DeepSeek publishes for both.
So is there any coding or agent task left where Pro is worth the premium? By DeepSeek's own numbers, very few. Pro keeps a narrow lead on one science-reasoning benchmark and tends to write shorter answers than the September Flash did. For agents, terminal work, and anything with images, the Flash is now the better and cheaper model.
The price gap
| Model | Input, peak | Output, peak | Cache hit, peak |
|---|---|---|---|
| V4.1 Flash 1017 | $0.30 | $1.20 | $0.006 |
| V4 Pro 0813 | $1.32 | $3.96 | $0.044 |
DeepSeek list prices per 1M tokens, from the DeepSeek pricing page. Off-peak is half price for both. AIHubMix lists Flash 1017 at $0.3098 / $1.2392 and Pro 0813 at $1.3944 / $4.1832 at peak.
Pro costs 4.4 times as much per input token, 3.3 times per output token, and over 7 times per cached token. Its concurrency limit on DeepSeek's API is also lower: 500 against the Flash's 2,500.
What DeepSeek's own numbers say
The Pro figures below come from Table 3 of DeepSeek's V4.1 Flash paper, all at max effort; the 1017 figures come from its model card. Both are DeepSeek's measurements and nobody has reproduced them. The paper's Pro column does not name the 0813 build, but its Terminal-Bench 2.1 score matches the 0813 card.
| Benchmark | V4 Pro | Flash (Sept) | Flash 1017 |
|---|---|---|---|
| Terminal-Bench 4.0 | 12.4 | 31.2 | 42.3 |
| DeepSWE v1.1 | 62.7 | 74.2 | 77.0 |
| ProgramBench | 15.5 | 20.3 | 29.1 |
| NL2Repo-Bench | 61.5 | 65.4 | 68.2 |
| Automation-Bench | 43.2 | 54.8 | 60.5 |
| Agents' Last Exam | 25.7 | 31.8 | 35.4 |
| HLE, text-only subset | 42.7 | 39.1 | 44.0 |
| GPQA Diamond | 92.4 | 90.9 | — |
A dash means DeepSeek has not published that score for 1017.
The agentic rows are not close. On Terminal-Bench 4.0, Pro scores 12.4 and 1017 scores 42.3; on DeepSWE, 62.7 against 77.0. Even the September Flash had already passed Pro on every coding and agent row. Artificial Analysis reached the same conclusion independently on its general index in September, calling V4.1 Flash DeepSeek's "new flagship" ahead of V4 Pro 0813.
Where Pro still has a case
The list is short, and each item is narrower than it looks.
- GPQA Diamond. Pro's 92.4 beats the September Flash's 90.9. DeepSeek hasn't published a 1017 score, so this is the one benchmark where Pro may still lead.
- Output length. Artificial Analysis found the September Flash used 62% more output tokens than Pro 0813 to run its index. 1017 was trained for shorter outputs, but no third party has re-measured it. Until someone does, a verbose Flash at a third of the price is still cheaper than a terse Pro.
- A stable, dated build. Pro 0813 has been in service since August, and DeepSeek's changelog says it will keep serving it. If your evaluations were tuned on Pro, it is the known quantity.
- Knowledge. On the text-only HLE subset, Pro (42.7) sits between the September Flash (39.1) and 1017 (44.0). That edge is gone.
Pro does not take images at all; 1017 does. If any step of your workflow reads a screenshot, a chart, or a UI, that alone decides it.
The same task on each
No published test gives per-task token counts for either model, so this example uses stated assumptions. An agent bug-fix session makes 40 model calls, sends 1.2M input tokens in total with 90% served from cache, and produces 60K output tokens including thinking. Both models get the same token counts. At AIHubMix list prices:
| Model | Peak | Off-peak |
|---|---|---|
| V4.1 Flash 1017 | $0.118 | $0.059 |
| V4 Pro 0813 | $0.469 | $0.233 |
Arithmetic for Pro at peak: 0.12M × $1.3944 + 1.08M × $0.0465 + 0.06M × $4.1832 = $0.469.
At equal token counts, Pro costs about four times as much for the same session. To break even, Pro would have to finish the job with roughly a quarter of the Flash's tokens. The gap Artificial Analysis measured is much smaller: the September Flash used 1.62 times Pro's output tokens, so Pro needed about 62% as many, not 25%. And on the benchmarks, the cheaper model is also the one more likely to finish the task.
Who still needs Pro
- Keep Pro if you have a production pipeline validated on Pro 0813 and no time to re-run evaluations this month, or a science-QA workload where the GPQA gap has shown up in your own tests.
- Move to the Flash for coding agents, terminal automation, tool-calling pipelines, anything with images, and anything high-volume where concurrency matters.
- Test before moving long-form writing and analysis where you have relied on Pro's shorter, denser answers. 1017 is shorter than the September Flash, but nobody has measured it against Pro yet.
How to test the switch on AIHubMix
Both models are on AIHubMix under one key: deepseek-v4.1-flash-1017 and deepseek-v4-pro-0813. Run your own evaluation set through each, compare pass rates, and check the cost of each run in the request logs. The AIHubMix compare page shows their specs and prices side by side.
import os
from openai import OpenAI
client = OpenAI(api_key=os.environ["AIHUBMIX_API_KEY"], base_url="https://aihubmix.com/v1")
for model in ("deepseek-v4-pro-0813", "deepseek-v4.1-flash-1017"):
resp = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": "Find the bug: def mean(xs): return sum(xs) / len(xs) - 1"}],
max_tokens=16000,
)
print(model, resp.usage.completion_tokens, resp.choices[0].message.content[:200])
Comparing completion_tokens across your own prompts is the quickest way to see whether 1017's shorter outputs hold up against Pro on your workload.
FAQ
Is DeepSeek V4.1 Flash 1017 better than V4 Pro?
On every coding and agent benchmark DeepSeek publishes for both, yes, often by a wide margin: Terminal-Bench 4.0 is 42.3 against 12.4, and DeepSWE is 77.0 against 62.7. These are DeepSeek's own figures.
How much more does V4 Pro cost?
At DeepSeek's peak prices, Pro costs $1.32 input and $3.96 output per 1M tokens against the Flash's $0.30 and $1.20, which is 3.3 times on output. Off-peak halves both.
Where is V4 Pro still better?
Pro scores 92.4 on GPQA Diamond against the September Flash's 90.9, and no 1017 score has been published. Artificial Analysis also measured the September Flash using more output tokens than Pro.
Does V4 Pro support images?
No. DeepSeek's pricing page lists vision for the Flash only.
Will DeepSeek retire V4 Pro?
DeepSeek said it will keep serving V4 Pro and has announced no retirement date.
Can 1017's shorter outputs make it cheaper than Pro per task?
It is already cheaper per task at equal token counts, by about four times in the example above. Shorter outputs widen that gap further.
Should I move a production Pro pipeline today?
Run your evaluation set on both first. If 1017 matches Pro's pass rate, the move cuts cost and adds headroom on concurrency.
Keep reading: the DeepSeek V4.1 Flash 1017 series
- Moving from Pro means moving to a model DeepSeek updates in place; the upgrade post explains what changed in 1017: DeepSeek V4.1 Flash 1017: Same Price, Shorter Outputs. Should You Switch?
- Switching a Pro integration to the Flash? The migration guide covers the thinking-mode 400 and the billing rules that apply to both: Migrating to DeepSeek V4.1 Flash 1017: The reasoning_content 400, the Silent Alias Swap, and Peak-Hour Billing
- If the Flash replaces Pro for you, the next question is how it stacks up outside DeepSeek's lineup: DeepSeek V4.1 Flash 1017 vs Claude Haiku 5.5: Which Cheap Model for Terminal Coding Agents?
Sources
- DeepSeek models and pricing (DeepSeek)
- DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression (DeepSeek-AI, arXiv)
- DeepSeek V4.1 Flash 1017 vs DeepSeek V4 Pro 0813 (AIHubMix)
- DeepSeek V4.1 Flash (max): Intelligence, Performance & Price Analysis (Artificial Analysis)
- DeepSeek V4.1 Flash 1017 Tested: Specs, Benchmarks & Price (Kingy.ai)
