DeepSeek V4.1 Flash vs DeepSeek V4.1 Flash 1017
Different efforts: non-thinking · max · Compare all at high
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
DeepSeek V4.1 Flash47.8
DeepSeek V4.1 Flash 101766.6
GDPval-AA (work tasks) 2.1 GDPval-AA (work tasks) 2.1: Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work
DeepSeek V4.1 Flash42.1
DeepSeek V4.1 Flash 101762.3
Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite
DeepSeek V4.1 Flash5.6
DeepSeek V4.1 Flash 101733.8
AA-Briefcase 1.1 · Elo AA-Briefcase 1.1 · Elo: Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating
DeepSeek V4.1 Flash1,117.6
DeepSeek V4.1 Flash 10171,570.4
AA-LCR (long context) 1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens)
DeepSeek V4.1 Flash55.3
DeepSeek V4.1 Flash 101782.7
FrontierCode 1.1 FrontierCode 1.1: Agents implement changes in real repositories, graded against maintainer-defined requirements
DeepSeek V4.1 Flash · max44
DeepSeek V4.1 Flash 101743.7
AA Analyst Agent AA Analyst Agent: Data-analyst tasks: the model explores data with tools and reports findings
DeepSeek V4.1 Flash · max35
DeepSeek V4.1 Flash 1017—
ITBench SRE ITBench SRE: IT incident tasks: diagnose and fix broken systems like a site-reliability engineer
DeepSeek V4.1 Flash · max46.9
DeepSeek V4.1 Flash 1017—
Terminal-Bench Science 0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal
DeepSeek V4.1 Flash · max9
DeepSeek V4.1 Flash 1017—
DeepSWE 1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition
DeepSeek V4.1 Flash—
DeepSeek V4.1 Flash 1017 · max53.3
Harvey LAB-AA 1.1 Harvey LAB-AA 1.1: Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit
DeepSeek V4.1 Flash · max1.7
DeepSeek V4.1 Flash 1017—
Furniture Assembly Furniture Assembly: Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step
DeepSeek V4.1 Flash · max34.2
DeepSeek V4.1 Flash 1017—
| Benchmark | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash 1017 |
|---|---|---|
| AutomationBench-AA | 47.8 | 66.6 |
| GDPval-AA (work tasks) 2.1 | 42.1 | 62.3 |
| Terminal-Bench 4.0 | 5.6 | 33.8 |
| AA-Briefcase 1.1 · Elo | 1,117.6 | 1,570.4 |
| AA-LCR (long context) 1.1 | 55.3 | 82.7 |
| FrontierCode 1.1 | 44 · max | 43.7 |
| AA Analyst Agent | 35 · max | — |
| ITBench SRE | 46.9 · max | — |
| Terminal-Bench Science 0.1 | 9 · max | — |
| DeepSWE 1.1 | — | 53.3 · max |
| Harvey LAB-AA 1.1 | 1.7 · max | — |
| Furniture Assembly | 34.2 · max | — |
Specs
| Spec | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash 1017 |
|---|---|---|
| Input, per 1M | !$0.30 | $0.30 |
| Output, per 1M | !$1.20 | $1.20 |
| Cost per task? | $0.22 | $0.26 |
| Speed, tok/s | 201 | 167 |
| Context | 1M | 1M |
| Weights | Open | Open |
| Released | Sep 2026 | Oct 2026 |
| Memory (4-bit) | — | 458 GB |
| Licence | — | MIT |
| Commercial use | — | Allowed |
| See images | ? | ✓ |
| Read PDFs | ? | — |
| Hear audio | ? | — |
| Watch video | ? | — |
| Call tools | ✓ | ✓ |
| Force a tool call | ? | ✓ |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
DeepSeek V4.1 Flash → DeepSeek V4.1 Flash 1017
Pass back reasoning_content · Chat Completions›
With thinking and tools, return all prior reasoning_content even on turns without tool calls or the API returns HTTP 400.
Disable thinking mode first · Chat Completions›
Required and named tool choices return HTTP 400 in thinking mode; disable thinking first.
Parameters have no effect›
In thinking mode, temperature, presence_penalty and frequency_penalty have no effect.
Values below 0.95›
Flash clamps top_p below 0.95 to 0.95 while thinking and ignores it in non-thinking mode, using 1.0.
Stateless API · Responses›
Responses does not support previous_response_id or conversation; preserve conversation history in your client.
DeepSeek V4.1 Flash 1017 → DeepSeek V4.1 Flash