Enter opens the highlighted result. Esc closes.

Codestral 2508 vs DeepSeek V3.2

Codestral 2508 and DeepSeek V3.2 cost about the same per output token ($0.90 vs $0.88 per million). Updated October 21, 2026; 6 benchmarks have a result for both.

Replace Codestral 2508 Replace DeepSeek V3.2
Codestral 2508Mistral AI
Coding
No data
Overall
No data
Output $/M
$0.90
0
DeepSeek V3.2DeepSeek
Coding
30.0
Overall
37.9
Output $/M
$0.88
6
vs boards won of 6 shared

Codestral 2508 or DeepSeek V3.2: which to pick

Only DeepSeek V3.2 has a coding score, 30.0 on the Vibe Coding Index; Codestral 2508 has none, so the two cannot be ranked on coding.

Pick Codestral 2508 for a larger context window (256K against 164K tokens).

Model ACodestral 2508Mistral AINot benchmarkedNo coding benchmarkNo cost per solved taskModel BDeepSeek V3.2DeepSeek30.0/1003 independent benchmarks$0.99 per solved task
Shared results
6
Codestral 2508 only
2
DeepSeek V3.2 only
69
Like-for-like categories
1 of 9

Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.

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
Codestral 2508 not benchmarked, DeepSeek V3.2 30.0 from 3 benchmarks.
Widest category
Web design, 18.1 points apart on average over 6 shared benchmarks.
Widest single gap
Design Arena: game dev: Codestral 2508 968 against DeepSeek V3.2 1132, 20.5 points apart on a 0 to 100 scale.
Output price
Codestral 2508 $0.90 against DeepSeek V3.2 $0.88 per million output tokens, 1.0x apart.
Context window
Codestral 2508 takes 256K tokens, DeepSeek V3.2 164K.

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
CategoryCodestral 2508BarsDeepSeek V3.2Ties
Web design06
All shared boards06
CodingDirectional onlySix of these boards feed the Vibe Coding Index15 boards, none shared
Coding benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
SWE-bench VerifiedNot measured70.0%Only DeepSeek V3.2 measured
SWE-bench MultilingualNot measured59.0%Only DeepSeek V3.2 measured
LMArena WebDevNot measured1362Only DeepSeek V3.2 measured
AA Coding IndexNot measured44.2Only DeepSeek V3.2 measured
LiveCodeBench (Vals AI)Not measured80.7%Only DeepSeek V3.2 measured
SWE-bench Verified (Vals AI)Not measured67.6%Only DeepSeek V3.2 measured
Vibe Code Bench (Vals AI)Not measured5.1%Only DeepSeek V3.2 measured
LiveCodeBench (Artificial Analysis run)Not measured86.2%Only DeepSeek V3.2 measured
Terminal-Bench Hard (Artificial Analysis run)Not measured35.6%Only DeepSeek V3.2 measured
Terminal-BenchNot measured39.6%Only DeepSeek V3.2 measured
ALE-Bench137.8Not measuredOnly Codestral 2508 measured
LMArena Text: codingNot measured1453Only DeepSeek V3.2 measured
LMArena Text: software and ITNot measured1449Only DeepSeek V3.2 measured
LMArena WebDev: ReactNot measured1350Only DeepSeek V3.2 measured
LMArena WebDev: HTMLNot measured1371Only DeepSeek V3.2 measured
Neither model has a result on the other 24 boards here
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
Intelligence benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
AA Intelligence IndexNot measured21.5Only DeepSeek V3.2 measured
Epoch Capabilities IndexNot measured146.1Only DeepSeek V3.2 measured
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
Agentic work benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
τ²-Bench TelecomNot measured90.6%Only DeepSeek V3.2 measured
APEX-AgentsNot measured21.3%Only DeepSeek V3.2 measured
Vending-Bench 2Not measured$1,034Only DeepSeek V3.2 measured
Neither model has a result on the other 38 boards here
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
Reasoning benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
GPQA DiamondNot measured80.3%Only DeepSeek V3.2 measured
GPQA Diamond (Artificial Analysis run)Not measured84.0%Only DeepSeek V3.2 measured
Humanity's Last Exam (Artificial Analysis run)Not measured24.6%Only DeepSeek V3.2 measured
ARC-AGINot measured57.0%Only DeepSeek V3.2 measured
ARC-AGI-2Not measured4.0%Only DeepSeek V3.2 measured
LMArena Text: hard promptsNot measured1434Only DeepSeek V3.2 measured
LMArena Text: expertNot measured1436Only DeepSeek V3.2 measured
LMArena Text: hard prompts in EnglishNot measured1446Only DeepSeek V3.2 measured
Kagi LLM Benchmark32.5%Not measuredOnly Codestral 2508 measured
Neither model has a result on the other 13 boards here
MathDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
Math benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
AIMENot measured84.6%Only DeepSeek V3.2 measured
MGSM (Vals AI)Not measured86.0%Only DeepSeek V3.2 measured
AA Math IndexNot measured92.0Only DeepSeek V3.2 measured
AIME 2025 (Artificial Analysis run)Not measured92.0%Only DeepSeek V3.2 measured
LMArena Text: mathNot measured1435Only DeepSeek V3.2 measured
LMArena Text: mathematical workNot measured1440Only DeepSeek V3.2 measured
Neither model has a result on the other 7 boards here
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
Knowledge benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
MMLU ProNot measured84.9%Only DeepSeek V3.2 measured
LegalBench (Vals AI)Not measured76.1%Only DeepSeek V3.2 measured
CaseLaw (Vals AI)Not measured55.4%Only DeepSeek V3.2 measured
CorpFin (Vals AI)Not measured51.0%Only DeepSeek V3.2 measured
TaxEval (Vals AI)Not measured68.2%Only DeepSeek V3.2 measured
MedQA (Vals AI)Not measured93.9%Only DeepSeek V3.2 measured
MMLU Pro (Artificial Analysis run)Not measured86.2%Only DeepSeek V3.2 measured
LMArena Text: scienceNot measured1446Only DeepSeek V3.2 measured
LMArena Text: medicine and healthcareNot measured1440Only DeepSeek V3.2 measured
LMArena Text: legal and governmentNot measured1425Only DeepSeek V3.2 measured
LMArena Text: business and financeNot measured1410Only DeepSeek V3.2 measured
Neither model has a result on the other 7 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index06
Web design benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
Design Arena: websites10131175DeepSeek V3.2
Design Arena: UI components10171149DeepSeek V3.2
Design Arena: game dev9681132DeepSeek V3.2
Design Arena: data viz10171158DeepSeek V3.2
Design Arena: 3D10201133DeepSeek V3.2
Design Arena: all code categories10071164DeepSeek V3.2
Design Arena: SVGNot measured1032Only DeepSeek V3.2 measured
Design Arena: ASCII artNot measured1083Only DeepSeek V3.2 measured
Neither model has a result on the other 1 board here
GeneralDirectional onlyPublic boards, shown as published, kept out of the index23 boards, none shared
General benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
IFBenchNot measured60.7%Only DeepSeek V3.2 measured
Long context reasoning (AA-LCR)Not measured73.3%Only DeepSeek V3.2 measured
DTBenchNot measured85.6%Only DeepSeek V3.2 measured
LMCANot measured28.8%Only DeepSeek V3.2 measured
CL-benchNot measured12.4%Only DeepSeek V3.2 measured
CL-bench LifeNot measured7.4%Only DeepSeek V3.2 measured
LMArena TextNot measured1425Only DeepSeek V3.2 measured
LMArena Text: instruction followingNot measured1413Only DeepSeek V3.2 measured
LMArena Text: creative writingNot measured1400Only DeepSeek V3.2 measured
LMArena Text: EnglishNot measured1439Only DeepSeek V3.2 measured
LMArena Text: non-EnglishNot measured1409Only DeepSeek V3.2 measured
LMArena Text: ChineseNot measured1455Only DeepSeek V3.2 measured
LMArena Text: GermanNot measured1413Only DeepSeek V3.2 measured
LMArena Text: JapaneseNot measured1374Only DeepSeek V3.2 measured
LMArena Text: KoreanNot measured1370Only DeepSeek V3.2 measured
LMArena Text: FrenchNot measured1429Only DeepSeek V3.2 measured
LMArena Text: SpanishNot measured1424Only DeepSeek V3.2 measured
LMArena Text: RussianNot measured1424Only DeepSeek V3.2 measured
LMArena Text: PolishNot measured1419Only DeepSeek V3.2 measured
LMArena Text: multi-turnNot measured1427Only DeepSeek V3.2 measured
LMArena Text: longer queriesNot measured1428Only DeepSeek V3.2 measured
LMArena Text: writing and languageNot measured1407Only DeepSeek V3.2 measured
LMArena Text: entertainment and mediaNot measured1395Only DeepSeek V3.2 measured
Neither model has a result on the other 23 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Codestral 2508 and DeepSeek V3.2
BenchmarkCodestral 2508DeepSeek V3.2Leader
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

Codestral 2508$0.0008Fits in one request
DeepSeek V3.2$0.0007Fits in one request

DeepSeek V3.2 costs less

Repository review

50K input + 3K output tokens

Codestral 2508$0.018Fits in one request
DeepSeek V3.2$0.017Fits in one request

DeepSeek V3.2 costs less

Agent loop

200K input + 10K output tokens

Codestral 2508$0.069Fits in one request
DeepSeek V3.2$0.067Exceeds 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 Codestral 2508 and DeepSeek V3.2
SpecificationMistral Codestral 2508DeepSeek V3.2
MakerMistral AIDeepSeek
WeightsNo public weights listedOpen weights
ReleasedAugust 1, 2025December 1, 2025
Context window256K164K
Input $/Mtok$0.30$0.29
Output $/Mtok$0.90$0.88
Coding indexNot benchmarked30.0
Independent measurements behind it03
Cost per solved taskNot computable$0.99
SourceOpenRouter listingOpenRouter listing

Questions

Is Codestral 2508 better than DeepSeek V3.2 for coding?

Only DeepSeek V3.2 has a coding score, 30.0 on the Vibe Coding Index; Codestral 2508 has none, so the two cannot be ranked on coding. Pick Codestral 2508 for a larger context window (256K against 164K tokens).

Which is cheaper, Codestral 2508 or DeepSeek V3.2?

Codestral 2508 costs $0.30 per million input tokens and $0.90 per million output tokens; DeepSeek V3.2 costs $0.29 and $0.88. Cost per solved task is known for DeepSeek V3.2 only.

Which has the bigger context window, Codestral 2508 or DeepSeek V3.2?

Codestral 2508, with 256K tokens against 164K.

How many benchmarks is the Codestral 2508 vs DeepSeek V3.2 comparison based on?

6 benchmarks have a result for both models as of October 21, 2026. Codestral 2508's coding index rests on 0 independent benchmarks and DeepSeek V3.2's on 3. Where only one model has a result, the page says so instead of naming a winner.

Compare next