Data as of October 19, 2026 · How the score is built
Released Dec 26, 2024128K context
DeepSeek V3
Decision readingDeepSeek V3 scores 31.9 out of 100 and ranks #176 of 226. This profile shows 9 source-displayable benchmark rows; its strongest eligible category is Instruction Following at #117. API pricing is $0.27 input and $1.1 output per million tokens, with cached input at $0.07.
Released Dec 26, 2024 — see all recent releases
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
31.9/100
field median 50#176 of 226 ranked models
Public
#176of 226
Verified #78 of 87
Price
$0.27input / $1.10 output
input median $0.75cached $0.070 · blended $0.69
Speed
37tok/s
field median 94 tok/sFirst token 1.25 s
Context
128Ktokens
field median 256,000Reported for this model; direct source link not stored
Strongest published evidence
Instruction Following ranks #117. A well-rounded choice across a range of tasks.
Validate before choosing
9 published rows leave some tracked benchmark slots empty. Knowledge is its lowest eligible category at #149.
Source-linked · 9 displayable benchmark rows
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticWeight 22%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| CodingRank #136 of 151Percentile 10thWeight 20%2 benchmarksReported | 14.7 | 2 benchmarks | Reported | |||
| ReasoningRank Not rankedWeight 17%1 benchmarkVerified | 25.5 | 1 benchmark | Verified | |||
| MultimodalWeight 12%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| KnowledgeRank #149 of 186Percentile 20thWeight 12%3 benchmarksMixed sources | 31.9 | 3 benchmarks | Mixed sources | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingRank #117 of 148Percentile 21stWeight 5%1 benchmarkReported | 39.6 | 1 benchmark | Reported | |||
| MathRank Not rankedWeight 5%2 benchmarksVerified | 1.7 | 2 benchmarks | Verified |
9 of 686 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- AgenticNot measured
- Coding0/2 verified
- Reasoning1/1 verified
- MultimodalNot measured
- Knowledge1/3 verified
- MultilingualNot measured
- Inst. Following0/1 verified
- Math2/2 verified
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
DeepSeek V3 category percentile values
- AgenticNot eligible
- Coding10th percentile
- ReasoningNot eligible
- MultimodalNot eligible
- Knowledge20th percentile
- MultilingualNot eligible
- Instruction following21st percentile
- MathNot eligible
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- AgenticNot ranked
- Coding#136/151
- ReasoningNot ranked
- MultimodalNot ranked
- Knowledge#149/186
- MultilingualNot ranked
- Inst. Following#117/148
- MathNot ranked
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| LiveCodeBenchLiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code | Score37.6% | Versus best verified row Best verified: Solar Pro 4 · 87.8% | Gap50.2 behind | WeightDisplay only | Reported |
| SWE-bench VerifiedSoftware Engineering Benchmark Verified | Score42% | Versus best verified row Best verified: Claude Opus 5 · 96% | Gap54 behind | WeightDisplay only | Reported |
Reasoning1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| LongBench v2 | Score48.7% | Versus best verified row Best verified: Beam 1.0 501B-A23B FP8 · 67.6% | Gap18.9 behind | WeightWeighted 25% | Provider exact |
Knowledge3 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMLU-ProMassive Multitask Language Understanding Professional | Score75.9% | Versus best verified row Best verified: Qwen3.7 Max · 89.6% | Gap13.7 behind | Weight6% ref. weight | Reported |
| GPQAGraduate-Level Google-Proof Q&A | Score59.1% | Versus best verified row Best verified: GPT-6 Astra · 96% | Gap36.9 behind | Weight3% ref. weight | Reported |
| SimpleQAMeasuring Short-Form Factuality in Large Language Models | Score24.9% | Versus best verified row Best verified: GPT-4.5 · 62.5% | Gap37.6 behind | WeightDisplay only | Provider exact |
Inst. Following1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| IFEvalInstruction-Following Eval | Score86.1% | Versus best verified row Best verified: LFM2.5-24B-A2B · 96.8% | Gap10.7 behind | WeightDisplay only | Reported |
Math2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| AIME 2024American Invitational Mathematics Examination 2024 | Score39.2% | Versus best verified row Best verified: GPT-OSS 120B · 95.8% | Gap56.6 behind | WeightDisplay only | Provider exact |
| MATH-500MATH-500 Problem Set | Score90.2% | Versus best verified row Best verified: Ternary Bonsai 2 27B · 98.8% | Gap8.6 behind | WeightDisplay only | Provider exact |
Bars run 0–100; the dark tick marks the best source-verified value
All 9 rowsWhat it costs to get this score
Published API price against the public score. The x-axis uses a log scale; the dashed path marks models that are not beaten by a cheaper, higher-scoring option. Price uses average of published input and output rates.
DeepSeek V3 · 31.9 score · $0.69 blended per million tokens
The chart opens on the current model. Scroll horizontally to inspect the full price axis.
Horizontal: blended price per million tokens, log scale · Vertical: public score
Published price history
This chart appears only when at least two dated first-party price records exist for the exact model. It shows rate changes, not an inferred cost trend.
2 dated first-party price records. The newest record lists $1.1 per million tokens.
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
snapshot · V3
DeepSeek V3 0324 · 36.5Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- Not published
- Context window
- 128K
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- Not sourced yet
- Output modalities
- Not sourced yet
- Parameters
- Not sourced yet
- Availability
- Not sourced yet
- Cloud regions
- Not tracked yet
- Lifecycle
- Superseded
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Published at $0.070 per million cached input tokens
- Self-host
- DeepSeek V3 needs ~640GB VRAM (8× NVIDIA H100 (80GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
DeepSeek V3 ranks #176 of 226 on the public leaderboard with a score of 31.86/100. Its source-verified position is #78 of 87.
DeepSeek V3 is a open weight model with a 128K context window. No explicit reasoning mode is documented in this profile.
DeepSeek V3 sits in the DeepSeek family with DeepSeek V3 0324. 9 of 686 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Instruction Following at #117, while its lowest eligible position is Knowledge at #149. a well-rounded choice across a range of tasks.
Last updated October 19, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does DeepSeek V3 perform overall in AI benchmarks?
DeepSeek V3 ranks #176 out of 226 models on the public BenchAlign leaderboard, with a score of 31.86/100. Its evidence status is Supported, and this profile shows 9 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is DeepSeek V3 good for knowledge and understanding?
DeepSeek V3 ranks #149 out of 186 eligible models for knowledge and understanding, with a public category score of 31.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is DeepSeek V3 good for coding and programming?
DeepSeek V3 ranks #136 out of 151 eligible models for coding and programming, with a public category score of 14.7/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is DeepSeek V3 good for mathematics?
DeepSeek V3 has source-displayable benchmark coverage for mathematics, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is DeepSeek V3 good for reasoning and logic?
DeepSeek V3 has source-displayable benchmark coverage for reasoning and logic, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is DeepSeek V3 good for instruction following?
DeepSeek V3 ranks #117 out of 148 eligible models for instruction following, with a public category score of 39.6/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is DeepSeek V3 open source?
DeepSeek V3 is an open-weight model from DeepSeek. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Which sibling models are related to DeepSeek V3?
DeepSeek V3 belongs to the DeepSeek family. Related tracked variants include DeepSeek V3 0324. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.
Does DeepSeek V3 have full benchmark coverage on BenchLM?
No. DeepSeek V3 currently has 25 source-displayable rows across 686 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of DeepSeek V3?
DeepSeek V3 has a reported context window of 128K in the exact-model catalog record. The value stays visible, but the profile marks its source link as unavailable instead of presenting it as directly documented. Maximum output length remains separate because providers often publish a different limit.