aiintelligence

DeepSeek V4 Prolatest from deepseek

It is designed to help users with advanced reasoning and computer coding tasks.

deepseek’s newest large language model, able to hold about 1049k tokens of context (roughly 786k words) in one conversation, working across text->text. You reach it through an API or through tools built on it. Built for developers and professionals who need help with coding and complex reasoning. deepseek describes it as: “DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window.” source ↗

Our tested score
86/100 · Meetings & notes
12 tasks, marked blind by 3 judges · 28 Aug 2026 · receipts
78/100 · Property & lettings
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
90/100 · HR & employment
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
70/100 · Bookkeeping & accounts
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
81/100 · Workflow automation
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
79/100 · Research agents
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
78/100 · Code review
12 tasks, marked blind by 3 judges · 18 Aug 2026 · receipts
89/100 · Humanising AI text
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
86/100 · Book writing
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
95/100 · Health questions
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
83/100 · Legal questions
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
97/100 · Emotional support
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
82/100 · Travel planning
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
92/100 · Translation
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
93/100 · Creative writing
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
98/100 · Research
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
86/100 · CV writing
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
91/100 · Presentations
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
63/100 · Vibe coding
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
88/100 · Job applications
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
94/100 · Social posts
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
97/100 · Flashcards
12 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
83/100 · Coding
18 tasks, marked blind by 3 judges · 16 Aug 2026 · receipts
85/100 · Study notes
12 tasks, marked blind by 3 judges · 14 Aug 2026 · receipts
93/100 · Customer replies
12 tasks, marked blind by 3 judges · 14 Aug 2026 · receipts
96/100 · Maths
12 tasks, marked blind by 3 judges · 14 Aug 2026 · receipts
93/100 · Emails
12 tasks, marked blind by 3 judges · 14 Aug 2026 · receipts
100/100 · Extraction
12 tasks, marked blind by 3 judges · 13 Aug 2026 · receipts
95/100 · Summarising
12 tasks, marked blind by 3 judges · 13 Aug 2026 · receipts
82/100 · Essays
12 tasks, marked blind by 3 judges · 13 Aug 2026 · receipts
93/100 · Spreadsheets
12 tasks, marked blind by 3 judges · 13 Aug 2026 · receipts
88/100 · Best-value API
30 tasks, marked blind by 3 judges · 11 Aug 2026 · receipts
API price (per 1M in/out)
$0.9245 / $1.8489
OpenRouter API21 Sept 2026
Community Elo (text)
1451
LMArena leaderboard (CC-BY-4.0)14 Sept 2026
Battle record
0W – 1L
in our published battles
Forum sentiment — 90d to 12 Aug 2026
86%
developer-forum posts: 7 opinions in 91
stale — measured 40 days ago, not re-swept since
What people are posting (90 days)
Loudest post this quarter
“DeepSeek V4 Pro beats GPT-5.5 Pro on precision”
Battles fought
DeepSeek V4 Pro vs GPT-5.6 Terra · Best-value API4–7GPT-5.6 Terra wins

Compare DeepSeek V4 Pro against…

Claude Opus 4.8 · summarising documentsGPT-5.5 · summarising documentsGPT-5.6 Luna · summarising documentsGPT-5.6 Terra · summarising documentsGPT-5.3-Codex · summarising documentsGPT-5.6 Sol · summarising documentsClaude Sonnet 5 · summarising documentsClaude Opus 4.6 · summarising documentsClaude Fable 5 · summarising documentsGemini 3.1 Flash Lite · extracting data from textGPT-5.6 Terra · extracting data from textGrok 4.5 · extracting data from textGPT-5.6 Sol · extracting data from textGPT-5.3-Codex · extracting data from textGemini 3.5 Flash · extracting data from textQwen3.7 Max · extracting data from textGemini 3.1 Pro Preview · extracting data from textKimi K3 · extracting data from textGPT-5.6 Sol · writing emails
Every claim we hold (6)