Share promptsSign inExplore benchmarks

DeepSeek vs ChatGPT: AI benchmark scores compared (Oct 2026)

On the 14 public leaderboards where both have a reported score, DeepSeek’s best-ranked model scores higher on 3, ChatGPT’s on 11. Counting only the 8 leaderboards their sources still update: DeepSeek 2, ChatGPT 6.

In plain words

  • Multi-domain agent tasks: ChatGPT higher on 1 of 1 (Vals Index).
  • Legal: ChatGPT higher on all 3 (Legal Benchmarks: Contract Workflows, Legal Benchmarks: Data Extraction and Legal Research Bench).
  • Health & medical: DeepSeek higher on 1 of 2 (HealthBench (independent run)); ChatGPT higher on 1 of 2 (MedQA).
  • Finance: DeepSeek higher on 1 of 1 (Finance Agent).
  • Coding: DeepSeek higher on 1 of 2 (SWE-bench Verified); ChatGPT higher on 1 of 2 (LiveCodeBench).
  • Agents & tool use: ChatGPT higher on all 2 (Terminal-Bench 4.0 and Terminal-Bench 2.1).
  • Reasoning & knowledge: ChatGPT higher on 1 of 1 (GPQA Diamond).
  • Chat preference: ChatGPT higher on 1 of 1 (LMArena Text Arena).
  • Multilingual: ChatGPT higher on 1 of 1 (MGSM).
  • DeepSeek’s best-ranked entry is most often DeepSeek V4.1 Flash 1017 (3 of 14); ChatGPT’s is most often GPT-6 Astra (5 of 14).

DeepSeek here means DeepSeek’s models; ChatGPT means OpenAI’s GPT and o-series models (the open-weight GPT OSS models are left out). On each leaderboard we compare each side’s best-ranked model, so the model can differ by benchmark. A side only counts where it appears on the leaderboard (we keep each source’s top rows).

Results from published AI leaderboards. Compare models within each benchmark; scores from different tests aren’t directly comparable. How we collect these results

Multi-domain agent tasks

BenchmarkDeepSeekChatGPTHigher score
Vals IndexAccuracy · dated 21 Oct 202661.26%DeepSeek V4.1 Flash 1017 · #865.02%GPT-6 Astra · #5ChatGPT higherby 3.76 pts

Health & medical

BenchmarkDeepSeekChatGPTHigher score
HealthBench (independent run)HealthBench · checked 22 Oct 202660.9%DeepSeek V4.1 Flash (max) · #258.5%GPT-6.1 Sol (max) · #3DeepSeek higherby 2.4 pts
MedQAAccuracy · dated 16 Apr 2026 · no longer updated93.92%DeepSeek V3.2 (Thinking) · #2096.52%o1 · #1ChatGPT higherby 2.60 pts

Finance

BenchmarkDeepSeekChatGPTHigher score
Finance AgentAccuracy · dated 19 Oct 202665.43%DeepSeek V4.1 Flash · #761.82%GPT-6.1 Sol · #13DeepSeek higherby 3.61 pts

Coding

BenchmarkDeepSeekChatGPTHigher score
SWE-bench VerifiedScore · dated 1 Sep 2026 · no longer updated96.4%DeepSeek V4 Pro 0813 · #296.2%GPT-5.6 Sol · #3DeepSeek higherby 0.2 pts
LiveCodeBenchAccuracy · dated 1 Sep 2026 · no longer updated87.53%DeepSeek V4 Pro 0813 · #1387.99%GPT 5.2 Codex · #9ChatGPT higherby 0.46 pts

Agents & tool use

BenchmarkDeepSeekChatGPTHigher score
Terminal-Bench 4.0Score · dated 19 Oct 202631.82%DeepSeek V4.1 Flash 1017 · #1759.60%GPT-6 Astra · #3ChatGPT higherby 27.78 pts
Terminal-Bench 2.1Score · dated 28 Sep 2026 · no longer updated74.53%DeepSeek V4.1 Flash · #1887.27%GPT-6 Astra · #2ChatGPT higherby 12.74 pts

Reasoning & knowledge

BenchmarkDeepSeekChatGPTHigher score
GPQA DiamondAccuracy · dated 1 Sep 2026 · no longer updated92.42%DeepSeek V4 Pro 0813 · #1795.20%GPT-5.6 Sol · #2ChatGPT higherby 2.78 pts

Chat preference

BenchmarkDeepSeekChatGPTHigher score
LMArena Text ArenaArena score · dated 20 Oct 20261475 ±6DeepSeek-V4.1-Flash-20260910 (Max) · #361484 ±4GPT-5.6 Sol (xHigh) · #21ChatGPT higherby 9, within ±

Multilingual

BenchmarkDeepSeekChatGPTHigher score
MGSMAccuracy · dated 9 Jan 2026 · no longer updated92.25%DeepSeek R1 · #2094.00%GPT 5.2 · #6ChatGPT higherby 1.75 pts

Every score links to the leaderboard that published it. Sources: Vals AI ↗ · Legal Benchmarks ↗ · Artificial Analysis ↗ · LMArena ↗. ChatJudge does not run these tests.

Test DeepSeek and ChatGPT on your own work

Run the same prompt in both and compare the answers. Prompts for ChatGPT →

ReviewsClaudeCompare two documents clause by clauseChecking a filled-in contract or proposal against its template.View prompt ProposalsChatGPTClarify commercial termsPricing, commission or partnership fees are being misread.View prompt
ResearchAny modelTurn messy text into a tableYou extract structured data from messy text. I will give you text and the fields I want. Return…View prompt AI Native SolutionsAI Native Solutions @ainativesolReport or remove
MarketingClaudeCustomer interviews to evidence-backed landing-page outlineFirst, extract exact quotes about desired outcomes, buying objections, and evidence customers ask…View prompt Nastation | AI for Client Work @NastaitonReport or remove
ResearchGeminiInvestor/stock due-diligenceAct as a forensic equity analyst. Conduct a deep research analysis for [COMPANY / TICKER] based on…View prompt MasterYoda @MasterYoda1991Report or remove
PricingClaudeUnit economics checkTesting whether your pricing survives your real usage costs.View prompt

Questions people ask

Is DeepSeek better than ChatGPT?

On the 14 public leaderboards where both have a reported score, DeepSeek’s best-ranked model scores higher on 3, ChatGPT’s on 11. Counting only the 8 leaderboards their sources still update: DeepSeek 2, ChatGPT 6. Which is better depends on the task: the leaders differ by benchmark, and scores from different tests are not directly comparable.

Is DeepSeek or ChatGPT better for multi-domain agent tasks?

Vals Index: ChatGPT 65.02% vs DeepSeek 61.26%. Each figure is the best-ranked model from each side on that leaderboard, as published by its source.

Is DeepSeek or ChatGPT better for legal work?

Legal Benchmarks: Contract Workflows: ChatGPT 25.8% vs DeepSeek 22.6%; Legal Benchmarks: Data Extraction: ChatGPT 56.7% vs DeepSeek 50.0%; Legal Research Bench: ChatGPT 48.08% vs DeepSeek 45.19%. Each figure is the best-ranked model from each side on that leaderboard, as published by its source.

Is DeepSeek or ChatGPT better for health & medical questions?

HealthBench (independent run): DeepSeek 60.9% vs ChatGPT 58.5%; MedQA: ChatGPT 96.52% vs DeepSeek 93.92%. Each figure is the best-ranked model from each side on that leaderboard, as published by its source.

Is DeepSeek or ChatGPT better for finance?

Finance Agent: DeepSeek 65.43% vs ChatGPT 61.82%. Each figure is the best-ranked model from each side on that leaderboard, as published by its source.

Is DeepSeek or ChatGPT better for coding?

SWE-bench Verified: DeepSeek 96.4% vs ChatGPT 96.2%; LiveCodeBench: ChatGPT 87.99% vs DeepSeek 87.53%. Each figure is the best-ranked model from each side on that leaderboard, as published by its source.

How recent is this comparison?

It is rebuilt from our benchmark data, last checked on 22 Oct 2026. Each leaderboard’s own date is shown in the table. MedQA, SWE-bench Verified, LiveCodeBench, Terminal-Bench 2.1, GPQA Diamond and MGSM are no longer updated by the source, so newer models may be missing there.

Which models are being compared?

On each leaderboard we take the best-ranked DeepSeek model and the best-ranked ChatGPT model, so the model can change from one benchmark to the next. DeepSeek’s top entry is most often DeepSeek V4.1 Flash 1017 (3 of 14); ChatGPT’s is most often GPT-6 Astra (5 of 14). DeepSeek here means DeepSeek’s models; ChatGPT means OpenAI’s GPT and o-series models (the open-weight GPT OSS models are left out).