AIEvals

Models

DeepSeek V4.1 Flash

DeepSeek’s open-weight V4.1 Flash release, with independent coding and reasoning results.

DeepSeekReleased 10 September 2026Context 1M tokensOutput 384k tokensOpen weights$0.30 in / $1.20 out per Mtok

Family DeepSeek V457 benchmarks with a published result1 independent first place among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI51.32%22of 3323of 41Gemini 4 Argon17.58 pts
Artificial Analysis Intelligence IndexArtificial Analysis3929of 42no list readClaude Opus 5.519 pts
Terminal-Bench 4.0Artificial Analysis26.77%19of 40no list readClaude Sonnet 5.536.87 pts
Vibe Code Bench v1.1Vals AI84.74%14of 3814of 106Claude Sonnet 5.57.65 pts
Terminal-Bench Science 0.1Artificial Analysis9.05%15of 24no list readGPT-6 Astra54.28 pts
ITBench-AAArtificial Analysis46.92%9of 18no list readGPT-5.6 Sol9.29 pts
MLCR-AAArtificial Analysis22.78%18of 37no list readClaude Sonnet 5.552.22 pts
Vibe Code Bench 1–100Vals AI16.38%12of 2012of 20Claude Opus 5.513.98 pts
AutomationBench-AAArtificial Analysis68.89%4of 40no list readGemini 4 Argon8.62 pts
SciCodeArtificial Analysis51.85%33of 40no list readClaude Opus 5.515.05 pts
GDP.pdfArtificial Analysis12.80%34of 40no list readGPT-6.1 Sol18.20 pts
CritPtArtificial Analysis14.29%35of 42no list readGPT-5.6 Sol18.00 pts
AA-OmniscienceArtificial Analysis-5.338of 42no list readClaude Opus 5.551.72
AA-LCR v1.1Artificial Analysis84.00%7of 42no list readKimi K34.67 pts
Terminal-Bench 2.1Vals AI74.53%18of 3518of 75Claude Opus 5.513.11 pts
ProofBench v1.1Vals AI54.00%22of 3426of 46Claude Sonnet 5.546.00 pts
LegalBenchVals AI83.28%25of 2754of 149Claude Fable 55.28 pts
Harvey's Legal Agent BenchmarkVals AI6.67%18of 3724of 73Muse Spark 1.218.75 pts
SkillsBenchVals AI69.80%1of 201of 35DeepSeek V4.1 Flash—
MysteryMechanismVals AI21.17%17of 2117of 21GPT-6 Astra31.98 pts
BioMysteryBenchVals AI67.78%14of 1914of 21Claude Sonnet 5.513.33 pts
Code MigrationVals AI45.62%15of 3715of 71Claude Sonnet 5.524.21 pts
CyberBench v1.1Vals AI73.69%7of 318of 43GPT-6 Sol4.29 pts
EMBVals AI57.21%28of 3736of 69Claude Fable 5.119.46 pts
Finance Agent (v2)Vals AI53.48%24of 3729of 73Gemini 4 Argon11.92 pts
IOIVals AI40.28%28of 3130of 38Gemini 4 Argon59.72 pts
Legal Research BenchVals AI41.35%20of 3721of 72Muse Spark 1.3 Max13.94 pts
MedCodeVals AI41.17%32of 3555of 104Claude Opus 522.40 pts
MedScribeVals AI85.50%20of 3625of 106Claude Opus 5.55.93 pts
Public Benefits Bench v1.1Vals AI64.28%20of 2920of 45Claude Opus 512.65 pts
SAGEVals AI47.88%17of 3229of 90Claude Opus 4.78.22 pts
SRE BenchVals AI0.76%14of 1514of 16GPT-6 Astra56.11 pts
Tax Agent BenchVals AI62.46%22of 3726of 64Claude Fable 5.115.18 pts
GDPval-AA v2.1Artificial Analysis1,60016of 42no list readClaude Opus 5.5246.17
AA-Briefcase v1.1Artificial Analysis1,420.5624of 41no list readClaude Opus 5.5401.29
Humanity's Last ExamArtificial Analysis39.25%36of 42no list readClaude Opus 5.522.10 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%Vibe Code Bench v1.1%Terminal-Bench Science 0.1%
Claude Sonnet 5Anthropic51.77%3814.14%81.33%–
GPT-5.6 LunaOpenAI51.69%3711.62%77.06%3.33%
DeepSeek V4.1 FlashDeepSeek51.32%3926.77%84.74%9.05%
Gemini 3.7 FlashGoogle51.27%3913.64%70.39%–
GPT-6 LunaOpenAI51.22%3712.63%81.65%8.57%
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

What the hosted runtime reports

Ollama Cloud serves its own copy of this model, so its context cap and its prices are the ones a caller actually gets. Read 25 September 2026.

Parameters
763B parameters8B active per token in prefill, 16B in decode
Hosted context
1M tokens
Hosted price / Mtok
$0.30in · $0.006 cached · $1.20 out · off-peak $0.15 in, $0.60 out
Vision inputTool callingReasoning modeServed by the hostInput: Text, Image

The host charges less outside peak hours: $0.15 in, $0.003 cached and $0.60 out per million tokens. DeepSeek describes the same model as a 552B backbone with 8B parameters active per token during prefill and 16B during decode.

ollama.com/library/deepseek-v4.1-flash

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
ITBench-AAIT operations46.92%Artificial Analysisnot published
MLCR-AAMedical reasoning22.78%Artificial Analysisnot published
Vibe Code Bench 1–100Application development16.38%Vals AInot published
Humanity's Last Exam (text-only, without tools)Multidisciplinary reasoningno published result
39.1%Text-only HLE (w/o tools)DeepSeek · 17 October 2026Rank 3 of 3 in this publisher table
Apex Bench v1.1Reasoningno published result
64.6%Apex Bench v1.1DeepSeek · 17 October 2026Rank 3 of 6 in this publisher table
AutomationBench (public split)Business workflowsno published result
54.8%AutomationBench publicReflection AI · 5 October 2026Rank 1 of 6 in this publisher table
AA-LCR (version not specified)Long-context reasoningno published result
84.0%AA-LCRReflection AI · 5 October 2026Rank 2 of 8 in this publisher table
AA-Omniscience (public split)Factual knowledgeno published result
6.6AA Omniscience (index) - public split only (our results)Reflection AI · 5 October 2026Rank 5 of 5 in this publisher table
AutomationBench-AAWorkflow automation68.89%Artificial Analysisnot published
SciCodeScientific coding51.85%Artificial Analysis
52.0%SciCodeReflection AI · 5 October 2026Rank 4 of 7 in this publisher table
GDP.pdfDocument reasoning12.80%Artificial Analysisnot published
CritPtScientific reasoning14.29%Artificial Analysis
14.3%CriPT AAReflection AI · 5 October 2026Rank 6 of 8 in this publisher table
AA-OmniscienceFactual knowledge-5.3Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning84.00%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding26.77%Artificial Analysis
19.70%Vals AI
31.2%Terminal-Bench 4.0DeepSeek · 17 October 2026Rank 5 of 6 in this publisher table
31.2%Terminal-Bench 4.0 (Pass@1)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
Terminal-Bench 2.1Agentic coding74.53%Vals AI
90.6%Terminal Bench v2.1Reflection AI · 5 October 2026Rank 1 of 8 in this publisher table
90.6%Terminal-Bench 2.1 (Pass@1)DeepSeek · 10 September 2026Rank 1 of 7 in this publisher table
GPQA DiamondScience reasoningno published result
90.9%GPQA DiamondReflection AI · 5 October 2026Rank 5 of 8 in this publisher table
90.9%GPQA Diamond (Pass@1)DeepSeek · 10 September 2026Rank 5 of 7 in this publisher table
MMMU ProMultimodal reasoning
76.99%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding84.74%Vals AInot published
Terminal-Bench Science 0.1Agentic science9.05%Artificial Analysis
0.00%Vals AI
not published
ProgramBenchAgentic codingno published result
20.3%ProgramBenchDeepSeek · 17 October 2026Rank 4 of 6 in this publisher table
20.3%ProgramBench (Almost@1)DeepSeek · 10 September 2026Rank 3 of 6 in this publisher table
ProofBench v1.1Mathematics54.00%Vals AInot published
LegalBenchProfessional knowledge83.28%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge6.67%Vals AInot published
SkillsBenchAgentic tooling69.80%Vals AInot published
MysteryMechanismScientific reasoning21.17%Vals AInot published
Vals IndexComposite51.32%Vals AInot published
BioMysteryBenchAgentic science67.78%Vals AInot published
Code MigrationAgentic coding45.62%Vals AInot published
CyberBench v1.1Security73.69%Vals AInot published
EMBFinancial modelling57.21%Vals AInot published
Finance Agent (v2)Financial analysis53.48%Vals AInot published
IOICompetitive programming40.28%Vals AInot published
Legal Research BenchProfessional knowledge41.35%Vals AInot published
MedCodeHealthcare administration41.17%Vals AInot published
MedScribeHealthcare administration85.50%Vals AInot published
Public Benefits Bench v1.1Public services64.28%Vals AInot published
SAGEEducation47.88%Vals AInot published
SRE BenchSecurity0.76%Vals AInot published
Tax Agent BenchProfessional knowledge62.46%Vals AInot published
AutomationBenchBusiness workflowsno published result
54.8%AutomationBench (v1.0.6)DeepSeek · 17 October 2026Rank 2 of 6 in this publisher table
54.8%AutomationBench (Pass@1)DeepSeek · 10 September 2026Rank 1 of 7 in this publisher table
DeepSWE v1.1Software engineeringno published result
74.2%DeepSWE v1.1DeepSeek · 17 October 2026Rank 2 of 6 in this publisher table
74.2%DeepSWE v1.1Reflection AI · 5 October 2026Rank 1 of 6 in this publisher table
74.2%DeepSWE v1.1 (Resolved)DeepSeek · 10 September 2026Rank 1 of 7 in this publisher table
GDPval-AA v2.1Knowledge work1,600Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work1,420.56Artificial Analysisnot published
ChartographyVisual chart recognitionno published result
78.9%Chartography (w/tools)DeepSeek · 17 October 2026Rank 4 of 5 in this publisher table
78.9%Chartography w/ tools (Pass@1)DeepSeek · 10 September 2026Rank 3 of 4 in this publisher table
Humanity's Last ExamMultidisciplinary reasoning39.25%Artificial Analysis
36.8%HLE (w/o tools)DeepSeek · 17 October 2026Rank 5 of 5 in this publisher table
39.1%HLE no toolsReflection AI · 5 October 2026Rank 5 of 8 in this publisher table
36.8%HLE (Pass@1)DeepSeek · 10 September 2026Rank 7 of 7 in this publisher table
63.9%HLE w/ tools (Pass@1)DeepSeek · 10 September 2026Rank 1 of 6 in this publisher table
Agents' Last ExamComputer useno published result
31.8%Agents' Last ExamDeepSeek · 17 October 2026Rank 2 of 6 in this publisher table
31.8%Agents' Last Exam (Pass@1)DeepSeek · 10 September 2026Rank 1 of 7 in this publisher table
Artificial Analysis Intelligence IndexComposite39Artificial Analysisnot published
Terminal-Bench 3.0Agentic codingno published result
30.0%Terminal-Bench 3.0 (Pass@1)DeepSeek · 10 September 2026Rank 3 of 7 in this publisher table
NL2Repo-BenchAgentic codingno published result
65.4%NL2Repo-BenchDeepSeek · 17 October 2026Rank 3 of 6 in this publisher table
65.4%NL2Repo-Bench (Score)DeepSeek · 10 September 2026Rank 2 of 7 in this publisher table
CyberGymSecurityno published result
88.1%CyberGym (Pass@1)DeepSeek · 10 September 2026Rank 1 of 6 in this publisher table
SEC-bench ProSecurityno published result
62.8%SEC-Bench Pro (Pass@1)DeepSeek · 10 September 2026Rank 2 of 4 in this publisher table
ExploitGymSecurityno published result
15.3%ExploitGym (Pass@1)DeepSeek · 10 September 2026Rank 3 of 6 in this publisher table
MathArena ApexMathematicsno published result
65.6%MathArena Apex (Pass@1)DeepSeek · 10 September 2026Rank 1 of 4 in this publisher table
Codeforces (rating)Competitive programmingno published result
3,471Codeforces (Rating)DeepSeek · 10 September 2026Rank 1 of 3 in this publisher table
BabyVisionMultimodalno published result
89.6%BabyVision (w/tools)DeepSeek · 17 October 2026Rank 3 of 5 in this publisher table
89.6%BabyVision w/tools (Pass@1)DeepSeek · 10 September 2026Rank 2 of 4 in this publisher table
ZeroBench (main set)Multimodalno published result
49.0%ZeroBench-main (w/tools)DeepSeek · 17 October 2026Rank 4 of 5 in this publisher table
49.0%ZeroBench-main w/tools (Pass@5)DeepSeek · 10 September 2026Rank 3 of 4 in this publisher table
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Sources

Read on 17 October 2026.