DeepSeekDeepSeek V3
DeepSeek V3 is a DeepSeek open-weights model, released 26 Dec 2024. Its best configuration ranks #551 of 785 on the BenchLeader Index at 44.0 ±3.2, in the lower half. The ± is the point: 285 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in human preference (52) and lowest in agents & tools (35). At $0.463 per million tokens blended it is cheaper than most ranked models. Output speed of 54 tokens per second puts it slower than most, with a first answer in 1.1 s. It has been superseded by DeepSeek V3 0324, which is the version to compare against today. Last measured 22 Oct 2026.
- Blended price
- $0.463/M
- $0.320 in · $0.890 out
- Output speed
- 54 tok/s
- measured by Artificial Analysis
- First answer
- 1.13 s
- first token 1.13 s
- Context
- 66k
- Full answer
- 10 s
- median, reasoning included
- Cost per run
- $0.020
- one full Intelligence Index run
- Released
- 26 Dec 2024
- Overall index44
- Reasoning40
- Coding47
- Agents & tools35
- Maths47
- Knowledge44
- Instruction following48
- Human preference52
- Long context39
- Composite39
Versions
DeepSeek has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| DeepSeek V3 0324 | 24 Mar 2025 | 53.5 | #287 |
| DeepSeek V3this page | 26 Dec 2024 | 44.0 | #551 |
How its index has moved
8 Sept 2026 to 20 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| GPQA Diamond | 56.5%#199 | Epoch AI Benchmarking Hub | |
| SimpleBench | 18.9%#82 | SimpleBench | |
| Arena Hard Prompts | 1350#215 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 55.7%#403 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 2.9%#607 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | 54.5%#111 | Vals AI | |
| Kagi LLM Benchmark | 52.3%#73 | Kagi LLM Benchmark | |
| GPQA Diamond (HELM)not in index | 53.8%#32 | HELM Capabilities | |
| CritPt | 0.0%#315 | Artificial Analysis | |
| SciPredictnot in index | 19.2%#8 | Scale AI SEAL | |
| ForecastBenchnot in index | 62.6%#53 | ForecastBench | |
| HellaSwagnot in index | 88.9%#4 | Epoch AI Benchmarking Hub | |
| BIG-Bench Hardnot in index | 87.5%#2 | Epoch AI Benchmarking Hub | |
| WinoGrandenot in index | 85.2%#6 | Epoch AI Benchmarking Hub | |
| ARC (AI2 Challenge)not in index | 95.3%#1 | Epoch AI Benchmarking Hub |
Coding
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| SciCode | 35.8%#188 | SciCode | |
| Arena Coding | 1387#206 | Arena (LMArena) | |
| SciCode (AA)not in index | 35.8%#200 | Artificial Analysis | |
| Aider Polyglot | 48.4%#27 | Aider polyglot leaderboard | |
| Terminal-Bench 4.0 (AA)not in index | 0.0%#152 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 16.9%#181 | Artificial Analysis |
Agents & tools
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Terminal-Bench Hard | 6.8%#249 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 22.8%#309 | Artificial Analysis | |
| τ³-Banking (AA) | 4.7%#188 | Artificial Analysis | |
| METR Time Horizonsnot in index | 47.4%#27 | METR | |
| AutomationBenchnot in index | 0.6%#203 | Artificial Analysis | |
| GDP.pdf (AA)not in index | 0.8%#210 | Artificial Analysis | |
| BrowseCompnot in index | 1.3%#13 | OpenAI |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| OTIS Mock AIME | 15.8%#226 | Epoch AI Benchmarking Hub | |
| MATH Level 5 | 64.8%#43 | Epoch AI Benchmarking Hub | |
| AIME (Vals) | 27.5%#71 | Vals AI | |
| MGSM | 92.2%#21 | Vals AI | |
| Omni-MATH (HELM) | 40.3%#30 | HELM Capabilities | |
| MATH 500not in index | 80.4%#36 | Vals AI | |
| Arena Maths | 1310#222 | Arena (LMArena) |
Knowledge
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-Omniscience | -41.6#329 | Artificial Analysis | |
| MMLU-Pro | 73.8%#109 | Vals AI | |
| LegalBench | 80.8%#82 | Vals AI | |
| CorpFin | 52.5%#96 | Vals AI | |
| TaxEval | 67.9%#100 | Vals AI | |
| MedQA | 80.9%#67 | Vals AI | |
| MMLU-Pro (HELM)not in index | 72.3%#35 | HELM Capabilities | |
| MMLUnot in index | 87.2%#3 | Epoch AI Benchmarking Hub | |
| TriviaQAnot in index | 82.9%#10 | Epoch AI Benchmarking Hub | |
| AA-Omniscience: accuracynot in index | 25.4%#228 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 10.0%#422 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| IFBench | 34.8%#313 | Artificial Analysis | |
| IFEval (HELM) | 83.2%#33 | HELM Capabilities | |
| WildBench (HELM) | 83.1%#19 | HELM Capabilities | |
| Arena Instruction Followingnot in index | 1343#196 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1358#192 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1348#169 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1373#177 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1375#183 | Arena (LMArena) |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 29.3%#397 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | 135.8#130 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 8.5#435 | Artificial Analysis | |
| HELM Capabilities meannot in index | 66.5%#31 | HELM Capabilities |
Safety & honesty
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| MASKnot in index | 44.5%#61 | Scale AI SEAL |
Price history
The price above, per 1M tokens, every day it changed.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0004 | – | 6.7 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0044 | – | 12.3 s |
| Code edit | 6,000 / 1,500 | $0.0033 | – | 29.1 s |
| Agentic coding session | 60,000 / 4,000 | $0.023 | – | 1.3 min |
| Structured extraction | 2,000 / 200 | $0.0008 | – | 4.9 s |
More from DeepSeek
See also
Questions people ask
- What are DeepSeek V3's benchmark scores?
- As of 24 Oct 2026, DeepSeek V3 scores 44.0 ±3.2 on the BenchLeader Index, strongest in human preference (52) and weakest in agents & tools (35). That rests on 28 independent benchmark results, and every score on this page links to its publisher.
- How does DeepSeek V3 rank?
- DeepSeek V3 is #551 of 785 ranked configurations on 24 Oct 2026. 285 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does DeepSeek V3 cost per 1M tokens?
- As of 24 Oct 2026, DeepSeek V3 costs $0.320 per 1M input tokens and $0.890 per 1M output tokens, $0.463 blended (three input tokens per output token). The list price last changed on 9 Sept 2026.
- When was DeepSeek V3 released?
- DeepSeek released DeepSeek V3 on 26 Dec 2024.
Data as of 24 Oct 2026. Compare with another model.
Cite as: BenchLeader, “DeepSeek V3: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/deepseek-v3.