Skip to content
VECTOR WIREAI INTELLIGENCE
UTC

DeepSeek-V4-Pro 0813 vs Mistral Large 4

DeepSeek · released

Wins 4 of 7 areas

Reasoning · Facts · Math · Following instructions

Mistral · released

Wins 2 of 7 areas

Coding · Long documents

DeepSeek-V4-Pro 0813 wins more areas, narrowly.Mistral Large 4 is cheaper and better at coding.

Scores updated · 31 tests both models report · How we compare

What it costs

Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.

DeepSeek-V4-Pro 0813$1.32 to read · $3.96 to write$5.28Price from Artificial Analysis
Mistral Large 4$0.70 to read · $2.10 to write$2.80Price from Artificial Analysis

Mistral Large 4 costs 47% less for the same work.

The biggest differences

The three tests each model wins by the widest margin. Scores are out of 100.

Where DeepSeek-V4-Pro 0813 pulls ahead

  • Short factual questions, answered correctlySimpleQA Verified+30.5points aheadDeepSeek-V4-Pro 081352.9Mistral Large 422.4
  • Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+23.3points aheadDeepSeek-V4-Pro 081349.1Mistral Large 425.8
  • Research-level physics problemsCritPt+8points aheadDeepSeek-V4-Pro 081318Mistral Large 410

Where Mistral Large 4 pulls ahead

  • Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+52.9points aheadMistral Large 458.1DeepSeek-V4-Pro 08135.2
  • Command-line tasks in a real terminalTerminal-Bench 2.1+11.5points aheadMistral Large 490.2DeepSeek-V4-Pro 081378.7
  • Complex command-line tasks across many fieldsTerminal-Bench 4.0+10.6points aheadMistral Large 424.7DeepSeek-V4-Pro 081314.1

Every test, side by side

All 31 tests both models report. The winning score is in its model's colour; marks a score checked independently.

CodingMistral Large 4

Full coding ranking

AgentsEven
  • Terminal-Bench 4.0Mistral Large 4 by 10.614.124.7+10.6
  • GDPValDeepSeek-V4-Pro 0813 by 1.347.546.2+1.3

Full agents ranking

ReasoningDeepSeek-V4-Pro 0813

Full reasoning ranking

FactsDeepSeek-V4-Pro 0813

Full facts ranking

MathDeepSeek-V4-Pro 0813

Full math ranking

Long documentsMistral Large 4
  • AA-LCRMistral Large 4 by 180.381.3+1

Full long documents ranking

Following instructionsDeepSeek-V4-Pro 0813

Full following instructions ranking

Other results14 tests, not counted

Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.

  • livebench_languageDeepSeek-V4-Pro 0813 by 32.582.149.6+32.5
  • JailbreakBench (cyber subset): safe-response rateMistral Large 4 by 32.464.997.3+32.4
  • StrongREJECT (cyber subset): refusal rateMistral Large 4 by 15.277.492.6+15.2
  • GDP (Surge AI)Mistral Large 4 by 6.211.417.6+6.2
  • AA-OmniscienceDeepSeek-V4-Pro 0813 by 6.10.8-5.3+6.1
  • AgentHarm (cyber subset): 100 minus harm scoreMistral Large 4 by 595100+5
  • CritPt v2 HardMistral Large 4 by 3.523.527+3.5
  • livebench_data_analysisDeepSeek-V4-Pro 0813 by 2.779.276.5+2.7
  • CritPt v2 (Hard Subset)Mistral Large 4 by 2.523.526+2.5
  • AA IntelligenceMistral Large 4 by 2.43638.4+2.4
  • HLE-DiamondDeepSeek-V4-Pro 0813 by 2.219.517.3+2.2
  • Terminal-Bench-Science 0.1DeepSeek-V4-Pro 0813 by 1.45.74.3+1.4
  • CyberGymDeepSeek-V4-Pro 0813 by 1.383.382+1.3
  • BeyondAIMEtie90.190.6tie

Questions people ask

Which is better, DeepSeek-V4-Pro 0813 or Mistral Large 4?

DeepSeek-V4-Pro 0813 wins four of the seven areas where both have results: reasoning, facts, math and following instructions. Mistral Large 4 wins coding and long documents, and costs 47% less. They are level on agents.

Which is better for coding?

Mistral Large 4. It wins 3 of the 4 coding tests both models report; DeepSeek-V4-Pro 0813 wins none, and 1 is a tie.

Which is cheaper?

DeepSeek-V4-Pro 0813 costs $1.32 per million input tokens and $3.96 per million output tokens; Mistral Large 4 costs $0.70 and $2.10. That makes Mistral Large 4 about 47% cheaper for the same work.

How do you compare the two?

We use the 31 benchmark tests both models have published scores on. The verdict counts the 17 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 14 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.