# Public code-agent benchmarks use DeepSeek Harness minimal mode, max effort, temperature=1.0, top_p=0.95.
# https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
name = "DeepSeek V4 Pro 0813"
description = "DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes"
family = "deepseek-thinking"
release_date = "2026-08-12"
last_updated = "2026-08-22"
attachment = false
reasoning = true
temperature = true
tool_call = true
structured_output = true
open_weights = true
license = "MIT"

[limit]
context = 1_000_000
output = 384_000

[modalities]
input = ["text"]
output = ["text"]

[[weights]]
label = "Hugging Face"
url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 42.7
metric = "score"
variant = "without tools"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 60
metric = "score"
variant = "with tools"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "Terminal-Bench"
score = 87.9
metric = "score"
harness = "DeepSeek Harness minimal mode"
variant = "max effort"
version = "2.1"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "NL2Repo"
score = 61.5
metric = "score"
harness = "DeepSeek Harness minimal mode"
variant = "max effort"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "CyberGym"
score = 83.3
metric = "score"
harness = "DeepSeek Harness minimal mode"
variant = "max effort"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "DeepSWE"
score = 62.7
metric = "score"
harness = "DeepSeek Harness minimal mode"
variant = "max effort"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "Toolathlon-Verified"
score = 74.1
metric = "score"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "Agents' Last Exam"
score = 25.7
metric = "score"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "AutomationBench"
score = 31.8
metric = "score"
dataset = "public"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "DSBench-FullStack"
score = 71.1
metric = "score"
dataset = "internal"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"

[[benchmarks]]
name = "DSBench-Hard"
score = 67.2
metric = "score"
dataset = "internal"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813"