Acropolis

DeepSeek V4.1 Flash 1017

DeepSeekGrade A. Evidence grade A of A to D: strongest.Open weightsReleased 17 Oct 2026Consistent across benchmarks
Compare with peers

Domain profile

Median within-board percentile by capability domain across boards with sufficient peer overlap.

  • Agentic (12 boards): 63 (scale 0 to 100)Agentic (12 boards)63
  • Coding (8 boards): 70 (scale 0 to 100)Coding (8 boards)70
  • Finance (4 boards): 61 (scale 0 to 100)Finance (4 boards)61
  • General text (1 boards): 79 (scale 0 to 100)General text (1 boards)79
  • Legal (3 boards): 68 (scale 0 to 100)Legal (3 boards)68
  • Science (2 boards): 85 (scale 0 to 100)Science (2 boards)85

Benchmark appearances

Best verified run per board, with confidence intervals and Pareto frontier status.

BenchmarkRankScoreCostStatusObserved
Tax Agent

Vals AI benchmarks · max

#6 / 38Tied #1
44.4%[37.3% to 51.4%]$0.46On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Vibe Code Bench

Vals AI benchmarks · max · OpenHands

#10 / 38
87.7%[82.3% to 93.1%]$0.58On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Code Migration

Vals AI benchmarks · max

#11 / 38
68.5%[61.7% to 75.3%]$1.57On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
LMArena WebDev

LMArena · Webdev HTML · max

#11 / 155
1,694[1,680 to 1,708]UnknownStandard15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · max · DeepSeek Harness

#13 / 145
42.7$1.84On frontier. Evidence grade A of A to D: strongest.18 Oct 2026
LMArena WebDev

LMArena · Image to webdev · max

#13 / 64
1,609[1,599 to 1,619]UnknownStandard15 Nov 2026
Legal Agent Benchmark (HLAB-Vals)

Vals AI benchmarks · max

#13 / 38
14.2%[8.7% to 19.6%]$0.61Standard15 Nov 2026
Vals Index

Vals AI benchmarks · max

#13 / 38
61.3%[59.2% to 63.3%]$0.66On frontier. Evidence grade A of A to D: strongest.15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · max · Pi

#15 / 145
42.2$0.83On frontier. Evidence grade A of A to D: strongest.18 Oct 2026
Legal Research

Vals AI benchmarks · max

#16 / 38
45.2%[38.4% to 52%]$0.70Standard15 Nov 2026
Finance Agent

Vals AI benchmarks · max

#18 / 38
62.9%[59% to 66.9%]$0.47Standard15 Nov 2026
LMArena WebDev

LMArena · Overall · max

#18 / 156
1,661[1,655 to 1,668]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev · max

#18 / 156
1,661[1,655 to 1,668]UnknownStandard15 Nov 2026
LMArena WebDev

LMArena · Webdev react · max

#18 / 140
1,673[1,666 to 1,681]UnknownStandard15 Nov 2026
Terminal-Bench 4

Vals AI benchmarks · max

#20 / 38
31.8%[26.7% to 37%]$0.54Standard15 Nov 2026
EMB

Vals AI benchmarks · max

#22 / 38
69.2%[64.5% to 73.9%]$0.33Standard15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · high · Pi

#23 / 145
40.4$0.64On frontier. Evidence grade A of A to D: strongest.18 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · high · DeepSeek Harness

#25 / 145
39.2$1.27Standard18 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · max · Claude Code

#26 / 145
39.1$1.01Standard18 Oct 2026
CursorBench

CursorBench · max · Cursor

#27 / 91
47.7%$0.98On frontier. Evidence grade A of A to D: strongest.Not stated
GPQA diamond

Epoch AI Benchmarking Hub · max

#27 / 338Tied #1
93.2%[90.1% to 96.3%]UnknownStandard19 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · max · OpenCode

#28 / 145
37.6$0.96Standard18 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · high · OpenCode

#29 / 145
36.6$0.79Standard19 Oct 2026
LMArena Agent

LMArena · Overall · max

#29 / 64
0[0 to 0]$0.14Standard15 Nov 2026
Bug Hunt Bench

Bug Hunt Bench · high · Claude Code

#31 / 145
36.4$0.82Standard17 Oct 2026
CursorBench

CursorBench · medium · Cursor

#31 / 91
46.3%$0.80On frontier. Evidence grade A of A to D: strongest.Not stated
CursorBench

CursorBench · high · Cursor

#35 / 91
45.6%$0.77On frontier. Evidence grade A of A to D: strongest.Not stated
Bug Hunt Bench

Bug Hunt Bench · low · DeepSeek Harness

#37 / 145
34.8$1.06Standard18 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · low · Claude Code

#38 / 145
34.4$0.71Standard17 Oct 2026
CursorBench

CursorBench · low · Cursor

#38 / 91
44.9%$0.60On frontier. Evidence grade A of A to D: strongest.Not stated
Bug Hunt Bench

Bug Hunt Bench · low · Pi

#42 / 145
33.8$0.46On frontier. Evidence grade A of A to D: strongest.18 Oct 2026
Bug Hunt Bench

Bug Hunt Bench · low · OpenCode

#44 / 145
32.8$0.66Standard19 Oct 2026
ARC-AGI-2

Epoch AI Benchmarking Hub · max

#55 / 299
78.1%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · low

#68 / 299
73.6%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · high

#78 / 299
68.6%UnknownStandardNot stated
ARC-AGI-2

Epoch AI Benchmarking Hub · none

#235 / 299
2.1%UnknownStandardNot stated

Reasoning effort scaling

Same benchmark and harness measured across multiple reasoning effort settings.

  • ARC-AGI-2

    Epoch AI Benchmarking Hub

    1 regression. Evidence grade D of A to D: weakest.
    none

    2.1

    low

    73.6

    high

    68.6

    max

    78.1

  • Bug Hunt Bench

    Bug Hunt Bench · Claude Code

    Monotone. Evidence grade A of A to D: strongest.
    low

    34.4

    $0.71

    high

    36.4

    $0.82

    max

    39.1

    $1.01

  • Bug Hunt Bench

    Bug Hunt Bench · DeepSeek Harness

    Monotone. Evidence grade A of A to D: strongest.
    low

    34.8

    $1.06

    high

    39.2

    $1.27

    max

    42.7

    $1.84

  • Bug Hunt Bench

    Bug Hunt Bench · OpenCode

    Monotone. Evidence grade A of A to D: strongest.
    low

    32.8

    $0.66

    high

    36.6

    $0.79

    max

    37.6

    $0.96

  • Bug Hunt Bench

    Bug Hunt Bench · Pi

    Monotone. Evidence grade A of A to D: strongest.
    low

    33.8

    $0.46

    high

    40.4

    $0.64

    max

    42.2

    $0.83

  • CursorBench

    CursorBench · Cursor

    1 regression. Evidence grade D of A to D: weakest.
    low

    44.9

    $0.60

    medium

    46.3

    $0.80

    high

    45.6

    $0.77

    max

    47.7

    $0.98

Closest head-to-head rivals

Direct pairwise win rates for DeepSeek V4.1 Flash 1017 across shared benchmark boards.