VerdictPal · editorial desk · 2026VerdictPal

Compare · Two models, head-to-head

Pick two models. See the diff.

Composite scores, benchmark ranks, capability grids, and pricing per million tokens. All sourced from the same data that powers the model atlas.

Tools Models

Mistral Large 3 vs DeepSeek V4 Pro Max

Model A

Model B

MODEL A

Mistral Large 3

Mistral · Mistral Large

Composite 48Intel 1620 rows6 sources

Mistral's December 2025 flagship MoE. 262K context, vision input, $0.50/$1.50 per 1M tokens. 675B total / 41B active parameters.

MODEL B

DeepSeek V4 Pro Max

DeepSeek · DeepSeek

Composite 74Intel 5614 rows10 sources

DeepSeek's April 2026 frontier open-weights model, leading the open-weights tier on SWE-bench Verified (80.6%) and Terminal-Bench 2.0 (67.9%) at $0.40 / $1.60 per 1M tokens. Cheapest frontier pricing on the public market.

Mistral Large 32stats won

Tied0stats

DeepSeek V4 Pro Max5stats won

Spec showdown

Capacity, price, freshness.

Mistral Large 3vsDeepSeek V4 Pro Max

DeepSeek V4 Pro Max leads on composite score (74 vs 48), with deeper benchmark coverage across more families. DeepSeek V4 Pro Max has the higher Vals/AA intelligence score (56 vs 16).

Composite score
Mistral Large 348DeepSeek V4 Pro Max74Winner on this row
Intelligence (Vals/AA)
Mistral Large 316DeepSeek V4 Pro Max56Winner on this row
Context window
Mistral Large 3262KWinner on this rowDeepSeek V4 Pro Max256K
Max output
Mistral Large 38KDeepSeek V4 Pro Max66KWinner on this row
Input price / 1M (lower wins)
Mistral Large 3$0.50DeepSeek V4 Pro Max$0.40Winner on this row
Output speed (AA default API)
Mistral Large 365 t/sWinner on this rowDeepSeek V4 Pro Max—
Time to first token (AA default API)
Mistral Large 30.67sDeepSeek V4 Pro Max—Winner on this row

Composite score

Intelligence (Vals/AA)

Context window

262K

256K

Max output

66K

Input price / 1M (lower wins)

$0.50

$0.40

Output speed (AA default API)

65 t/s

—

Time to first token (AA default API)

0.67s

—

Capability grid

What each model can do.

Modalities

Mistral Large 3DeepSeek V4 Pro Max

Text●●

Image●—

Audio——

Video——

Code●●

Tool use●●

Reasoning●●

Access

Mistral Large 3DeepSeek V4 Pro Max

API●●

Consumer app——

Open weights●●

Managed cloud●●

On-device——

Family radar

Top score per benchmark family.

Knowledge

Reasoning

Math

Coding

Agentic

Long context

Tool use

Benchmark ranks

Every shared benchmark, side by side.

Knowledge

MMLU#13/22 · #8/22

87.8%

89.7%

MMLU-Pro#31/34 · #27/34

72.8%

76.4%

Reasoning

GPQA Diamond#135/151 · #124/151

71.2%

75.6%

LiveBench#11/21 · #14/21

62.8%

61.9%

Coding

HumanEval#12/15 · #9/15

92.4% pass@1

93.4% pass@1

SWE-bench Verified#21/23 · #7/23

76.2%

80.6%

Agentic

Terminal-Bench#141/150 · #35/150

11.99%

67.9%

Long context

LongBench v2#11/19 · #14/19

50.8%

48.6%

Tool use

BFCL#8/20 · #16/20

80.1%

74.6%

MCP Atlas#8/10 · #6/10

70.4%

72.6%

Dossier facts

Dates, sources, fineprint.

Mistral Large 3DeepSeek V4 Pro Max

ProviderMistralDeepSeek

FamilyMistral LargeDeepSeek

Release date2025-12-022026-04-24

Knowledge cutoff2025-06-012026-02-01

Context window262K256K

Max output8K66K

Input price /1M$0.50$0.40

Output price /1M$1.5$1.6

Statussolidflagship

Composite score4874

Intelligence score1656

Intel sourceaavals

Benchmark rows2014

Source count610

Freshnessfreshfresh

Pricing checked2026-06-092026-06-01

Ecosystem

Which tools wrap each model.

Mistral Large 3

mistral le chat

DeepSeek V4 Pro Max

deepseek

Verdict

Which model to actually pick.

DeepSeek V4 Pro Max leads on composite score (74 vs 48), with deeper benchmark coverage across more families. DeepSeek V4 Pro Max has the higher Vals/AA intelligence score (56 vs 16).

Open Mistral Large 3's dossier Open DeepSeek V4 Pro Max's dossier