DeepSeek V4.1 Flash 1017 1017

DeepSeek · release date unknown

Open-weight (permissive)CNtext

Openness taxonomy

access, licence, and disclosure — never a single open/closed binary

Weights availability: Open-weight model; Unsloth shipped a llama.cpp implementation with dynamic GGUFs.
Licence: MIT
Licence category: permissive
Commercial use allowed: unknown
Derivatives allowed: unknown
Redistribution allowed: unknown
Field-of-use restrictions: unknown
Training code availability: unknown
Inference code availability: llama.cpp implementation available
Architecture disclosure: 552B-total/16B-active MoE; 1M-token context window
Training data disclosure: unknown

openness confidence

Artificial Analysis lists an MIT license.
Source describes it as an open-weight model.
Unsloth shipped a llama.cpp implementation with dynamic GGUFs.

Deployability

raw values as extracted — free text, never inferred from branding

Self-hosting: Local deployment implementation available
On-premise: unknown
Edge deployability: unknown
API availability: DeepSeek API and third-party providers including Fireworks, Baseten, DeepInfra, Cloudflare and Together
Regions available: unknown

Benchmarks

as reported — vendor-supplied unless otherwise noted

Artificial Analysis Intelligence Index 45GDPval-AA Elo 1746CritPt 24%AA-Omniscience non-hallucination rate 33%Code Arena WebDev 1670 pointsAgent Arena #11 overallVals Index v2.2 #8 overall; top open-weight modelLong-horizon browser use 75/100Terminal-Bench 4.0 42.3

National-security pathway & evidence

from the assessed source item, when one is linked

Impact: mediumConfidence: mediumNat-sec: indirectanalysis

Domains: tech_sovereignty

Pathway: Availability of an open-weight model that can be deployed locally may reduce dependence on hosted model APIs for some users.

Evidence limitations: This is a single commentary roundup combining third-party evaluations, vendor-reported figures and community testing; the source notes a harness change that confounds some comparisons.

[AINews] not much happened today

Linked research papers

from paper_model_links

Repair, Retry, and the Cost of Hindsight: Verifier Feedback in Formal Theorem Provingevaluates · 100% confidence
Rethinking GRPO for Large Reasoning Model Safety Alignmentevaluates · 95% confidence