# Benchmarks use instruct weights, temperature=1.0, top_p=0.95; not the base-model table.
# Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash#instruct-model
# Open weights (MIT): https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
name = "DeepSeek V4.1 Flash"
description = "DeepSeek V4.1 Flash model for reasoning and agentic coding"
family = "deepseek-flash"
release_date = "2026-09-10"
last_updated = "2026-09-10"
attachment = true
reasoning = true
temperature = true
tool_call = true
structured_output = true
knowledge = "2025-05"
open_weights = true
license = "MIT"

[limit]
context = 1_000_000
output = 384_000

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

[[weights]]
label = "Hugging Face"
url = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "GPQA Diamond"
score = 90.9
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "MathArena Apex"
score = 65.6
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 36.8
metric = "pass@1"
variant = "reasoning_effort=100"
dataset = "full set"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 39.1
metric = "pass@1"
variant = "reasoning_effort=100"
dataset = "text-only subset"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Codeforces"
score = 3471
metric = "rating"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 90.6
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 30.0
metric = "pass@1"
variant = "reasoning_effort=100"
version = "3.0"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 31.2
metric = "pass@1"
variant = "reasoning_effort=100"
version = "4.0"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 74.2
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "mini-swe-agent"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "ProgramBench"
score = 20.3
metric = "almost@1"
variant = "reasoning_effort=100"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "NL2Repo"
score = 64.0
metric = "score"
variant = "reasoning_effort=100"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "CyberGym"
score = 88.1
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "ExploitGym"
score = 15.3
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "AutomationBench"
score = 54.8
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Agents' Last Exam"
score = 31.8
metric = "pass@1"
variant = "reasoning_effort=100"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "SEC-Bench Pro"
score = 62.8
metric = "pass@1"
variant = "reasoning_effort=100"
harness = "Claude Code"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 63.9
metric = "pass@1"
variant = "reasoning_effort=100; with tools"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Chartography"
score = 78.9
metric = "pass@1"
variant = "reasoning_effort=100; with tools"
harness = "Claude Code"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "BabyVision"
score = 89.6
metric = "pass@1"
variant = "reasoning_effort=100; with tools"
harness = "Claude Code"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "ZeroBench"
score = 49.0
metric = "pass@5"
variant = "reasoning_effort=100; with tools"
harness = "Claude Code"
dataset = "main"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 69.8
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "Claude Code"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 88.0
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "Claude Code"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 65.6
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "Codex"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 84.1
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "Codex"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 65.5
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "OpenCode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 85.0
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "OpenCode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 66.2
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "Pi"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 86.1
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "Pi"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 90.3
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "mini-swe-agent"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 72.6
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "DeepSeek Harness minimal mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 70.5
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "DeepSeek Harness standard mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 85.8
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "DeepSeek Harness standard mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "DeepSWE"
score = 67.6
metric = "resolved"
variant = "reasoning_effort=100"
version = "1.1"
harness = "DeepSeek Harness PTC mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"

[[benchmarks]]
name = "Terminal-Bench"
score = 85.8
metric = "pass@1"
variant = "reasoning_effort=100"
version = "2.1"
harness = "DeepSeek Harness PTC mode"
source = "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"