Enter opens the highlighted result. Esc closes.

DeepSeek V3.2 vs Llama 3.3 70B Instruct

Updated October 21, 2026. 43 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace DeepSeek V3.2 Replace Llama 3.3 70B Instruct
DeepSeek V3.2DeepSeek
Coding
30.0
Overall
21.5
Output $/M
$0.88
43
Llama 3.3 70B InstructMeta Llama
Coding
No data
Overall
7.7
Output $/M
$0.72
0
vs boards won of 43 shared

Decision reading

Only DeepSeek V3.2 has a coding score, 30.0 on the Vibe Coding Index; Llama 3.3 70B Instruct has none, so the two cannot be ranked on coding.

Llama 3.3 70B Instruct has the edge where output tokens cost 1.2x less.

Model ADeepSeek V3.2DeepSeek30.0/1003 independent benchmarks$0.99 per solved taskModel BLlama 3.3 70B InstructMeta LlamaNot benchmarkedNo coding benchmarkNo cost per solved task
Shared results
43
DeepSeek V3.2 only
32
Llama 3.3 70B Instruct only
8
Like-for-like categories
7 of 9

Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.

Which one for your work

Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.

At a glance

Coding index
DeepSeek V3.2 30.0 from 3 benchmarks, Llama 3.3 70B Instruct not benchmarked.
Widest category
Agentic work, 64.0 points apart on average over 1 shared benchmark.
Widest single gap
AIME 2025 (Artificial Analysis run): DeepSeek V3.2 92.0% against Llama 3.3 70B Instruct 7.7%, 84.3 points apart on a 0 to 100 scale.
Output price
DeepSeek V3.2 $0.88 against Llama 3.3 70B Instruct $0.72 per million output tokens, 1.2x apart.
Context window
DeepSeek V3.2 takes 164K tokens, Llama 3.3 70B Instruct 131K.

Benchmark by benchmark

First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.

Boards won, where both have a result
CategoryDeepSeek V3.2BarsLlama 3.3 70B InstructTies
Coding50
Intelligence20
Agentic work10
Reasoning50
Math40
Knowledge50
General210
All shared boards430
CodingLike-for-likeSix of these boards feed the Vibe Coding Index50
Coding benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
SWE-bench Verified70.0%Not measuredOnly DeepSeek V3.2 measured
SWE-bench Multilingual59.0%Not measuredOnly DeepSeek V3.2 measured
LMArena WebDev1362Not measuredOnly DeepSeek V3.2 measured
AA Coding Index44.211.9DeepSeek V3.2
LiveCodeBench (Vals AI)80.7%Not measuredOnly DeepSeek V3.2 measured
SWE-bench Verified (Vals AI)67.6%Not measuredOnly DeepSeek V3.2 measured
Vibe Code Bench (Vals AI)5.1%Not measuredOnly DeepSeek V3.2 measured
LiveCodeBench (Artificial Analysis run)86.2%28.8%DeepSeek V3.2
Terminal-Bench Hard (Artificial Analysis run)35.6%3.0%DeepSeek V3.2
Terminal-Bench39.6%Not measuredOnly DeepSeek V3.2 measured
WeirdMLNot measured14.4%Only Llama 3.3 70B Instruct measured
LMArena Text: coding14531268DeepSeek V3.2
LMArena Text: software and IT14491273DeepSeek V3.2
LMArena WebDev: React1350Not measuredOnly DeepSeek V3.2 measured
LMArena WebDev: HTML1371Not measuredOnly DeepSeek V3.2 measured
Neither model has a result on the other 23 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
Intelligence benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
AA Intelligence Index21.57.7DeepSeek V3.2
Epoch Capabilities Index146.1127.3DeepSeek V3.2
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
Agentic work benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
τ²-Bench Telecom90.6%26.6%DeepSeek V3.2
APEX-Agents21.3%Not measuredOnly DeepSeek V3.2 measured
BALROGNot measured23.0%Only Llama 3.3 70B Instruct measured
Vending-Bench 2$1,034Not measuredOnly DeepSeek V3.2 measured
Neither model has a result on the other 35 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index50
Reasoning benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
GPQA Diamond80.3%Not measuredOnly DeepSeek V3.2 measured
GPQA Diamond (Artificial Analysis run)84.0%49.8%DeepSeek V3.2
Humanity's Last Exam (Artificial Analysis run)24.6%3.6%DeepSeek V3.2
GPQA Diamond (Epoch AI run)Not measured47.4%Only Llama 3.3 70B Instruct measured
ARC-AGI57.0%Not measuredOnly DeepSeek V3.2 measured
ARC-AGI-24.0%Not measuredOnly DeepSeek V3.2 measured
SimpleBenchNot measured19.9%Only Llama 3.3 70B Instruct measured
LMArena Text: hard prompts14341257DeepSeek V3.2
LMArena Text: expert14361225DeepSeek V3.2
LMArena Text: hard prompts in English14461277DeepSeek V3.2
Neither model has a result on the other 11 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index40
Math benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
AIME84.6%Not measuredOnly DeepSeek V3.2 measured
MGSM (Vals AI)86.0%Not measuredOnly DeepSeek V3.2 measured
AA Math Index92.07.7DeepSeek V3.2
AIME 2025 (Artificial Analysis run)92.0%7.7%DeepSeek V3.2
OTIS Mock AIMENot measured5.1%Only Llama 3.3 70B Instruct measured
MATH level 5Not measured41.6%Only Llama 3.3 70B Instruct measured
LMArena Text: math14351267DeepSeek V3.2
LMArena Text: mathematical work14401267DeepSeek V3.2
Neither model has a result on the other 5 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index50
Knowledge benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
MMLU Pro84.9%Not measuredOnly DeepSeek V3.2 measured
LegalBench (Vals AI)76.1%Not measuredOnly DeepSeek V3.2 measured
CaseLaw (Vals AI)55.4%Not measuredOnly DeepSeek V3.2 measured
CorpFin (Vals AI)51.0%Not measuredOnly DeepSeek V3.2 measured
TaxEval (Vals AI)68.2%Not measuredOnly DeepSeek V3.2 measured
MedQA (Vals AI)93.9%Not measuredOnly DeepSeek V3.2 measured
MMLU Pro (Artificial Analysis run)86.2%71.3%DeepSeek V3.2
MMLUNot measured86.3%Only Llama 3.3 70B Instruct measured
LMArena Text: science14461284DeepSeek V3.2
LMArena Text: medicine and healthcare14401275DeepSeek V3.2
LMArena Text: legal and government14251284DeepSeek V3.2
LMArena Text: business and finance14101243DeepSeek V3.2
Neither model has a result on the other 5 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
Web design benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
Design Arena: websites1175Not measuredOnly DeepSeek V3.2 measured
Design Arena: UI components1149Not measuredOnly DeepSeek V3.2 measured
Design Arena: game dev1131Not measuredOnly DeepSeek V3.2 measured
Design Arena: data viz1158Not measuredOnly DeepSeek V3.2 measured
Design Arena: 3D1133Not measuredOnly DeepSeek V3.2 measured
Design Arena: all code categories1164Not measuredOnly DeepSeek V3.2 measured
Design Arena: SVG1032Not measuredOnly DeepSeek V3.2 measured
Design Arena: ASCII art1082Not measuredOnly DeepSeek V3.2 measured
Neither model has a result on the other 1 board here
GeneralLike-for-likePublic boards, shown as published, kept out of the index210
General benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
IFBench60.7%47.1%DeepSeek V3.2
Long context reasoning (AA-LCR)73.3%15.7%DeepSeek V3.2
DTBench85.6%59.5%DeepSeek V3.2
LMCA28.8%17.5%DeepSeek V3.2
Fiction.LiveBenchNot measured33.3%Only Llama 3.3 70B Instruct measured
CL-bench12.4%Not measuredOnly DeepSeek V3.2 measured
CL-bench Life7.4%Not measuredOnly DeepSeek V3.2 measured
LMArena Text14251274DeepSeek V3.2
LMArena Text: instruction following14131242DeepSeek V3.2
LMArena Text: creative writing14001250DeepSeek V3.2
LMArena Text: English14391303DeepSeek V3.2
LMArena Text: non-English14091236DeepSeek V3.2
LMArena Text: Chinese14551217DeepSeek V3.2
LMArena Text: German14131251DeepSeek V3.2
LMArena Text: Japanese13741150DeepSeek V3.2
LMArena Text: Korean13701142DeepSeek V3.2
LMArena Text: French14291282DeepSeek V3.2
LMArena Text: Spanish14241269DeepSeek V3.2
LMArena Text: Russian14241252DeepSeek V3.2
LMArena Text: Polish14191259DeepSeek V3.2
LMArena Text: multi-turn14271280DeepSeek V3.2
LMArena Text: longer queries14281255DeepSeek V3.2
LMArena Text: writing and language14071255DeepSeek V3.2
LMArena Text: entertainment and media13951241DeepSeek V3.2
Neither model has a result on the other 22 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for DeepSeek V3.2 and Llama 3.3 70B Instruct
BenchmarkDeepSeek V3.2Llama 3.3 70B InstructLeader
Neither model has a result on 5 boards here

What each workload costs

Three fixed token mixes at each model's listed price, so per-token rates become a bill.

Chat turn

1K input + 500 output tokens

DeepSeek V3.2$0.0007Fits in one request
Llama 3.3 70B Instruct$0.0008Fits in one request

DeepSeek V3.2 costs less

Repository review

50K input + 3K output tokens

DeepSeek V3.2$0.017Fits in one request
Llama 3.3 70B Instruct$0.025Fits in one request

DeepSeek V3.2 costs less

Agent loop

200K input + 10K output tokens

DeepSeek V3.2$0.067Exceeds the context window
Llama 3.3 70B Instruct$0.097Exceeds the context window

DeepSeek V3.2 costs less

Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.

Specification differences

The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.

Specifications of DeepSeek V3.2 and Llama 3.3 70B Instruct
SpecificationDeepSeek V3.2Meta Llama 3.3 70B Instruct
MakerDeepSeekMeta Llama
WeightsOpen weightsOpen weights
ReleasedDecember 1, 2025December 6, 2024
Context window164K131K
Input $/Mtok$0.29$0.45
Output $/Mtok$0.88$0.72
Coding index30.0Not benchmarked
Independent measurements behind it30
Cost per solved task$0.99Not computable
SourceOpenRouter listingOpenRouter listing

Questions

Is DeepSeek V3.2 better than Llama 3.3 70B Instruct for coding?

Only DeepSeek V3.2 has a coding score, 30.0 on the Vibe Coding Index; Llama 3.3 70B Instruct has none, so the two cannot be ranked on coding. Llama 3.3 70B Instruct has the edge where output tokens cost 1.2x less.

Which is cheaper, DeepSeek V3.2 or Llama 3.3 70B Instruct?

DeepSeek V3.2 costs $0.29 per million input tokens and $0.88 per million output tokens; Llama 3.3 70B Instruct costs $0.45 and $0.72. Cost per solved task is known for DeepSeek V3.2 only.

Which has the bigger context window, DeepSeek V3.2 or Llama 3.3 70B Instruct?

DeepSeek V3.2, with 164K tokens against 131K.

How many benchmarks is the DeepSeek V3.2 vs Llama 3.3 70B Instruct comparison based on?

43 benchmarks have a result for both models as of October 21, 2026. DeepSeek V3.2's coding index rests on 3 independent benchmarks and Llama 3.3 70B Instruct's on 0. Where only one model has a result, the page says so instead of naming a winner.

Compare next