DeepSeek V4.1 Flash hands-on tests and comparisons

Tested with the prompt "create svg of red lamborghini", the 16B-active Flash model produced full SVG geometry on the first try, which the author found surprising.

Official update: Search and Fetch APIs open to developers, V4.1-Flash moves to 1017 at unchanged pricing, with faster tasks, shorter outputs and DeepSeek Harness tuning.

vLLM reports DeepSeek-V4.1-Flash serving 830K tok/s per GPU and 1.3B tokens per dollar TCO on SemiAnalysis AgentX, with 5x agentic throughput since day one.

DeepSWE 77.0 vs 74.0, Automation-Bench 60.5 vs 50.3, Agents' Last Exam 35.4 vs 28.6 and Apex 72.1 vs 65.8, though Terminal-Bench and HLE still trail; the post stresses this is the smallest 552B MoE model with open weights.

The post cites 42.6 on HLE, 42.3 on Terminal-Bench and 77.0 on DeepSWE, saying the Flash model beats GPT-5.6 Sol on several benchmarks despite not being the full version.

Run through DeepSeek Harness at max reasoning, it produced a single file with over 2 million voxels in about 25 minutes for roughly $5 in API cost.

The post weighs market impact of V4.1 Flash-1017: MIT open weights, $0.15 input and $0.60 output per 1M tokens off-peak, recalling Nvidia's 17% drop after the last DeepSeek weekend release.

Terminal-Bench 4.0 rose from 31.2 to 42.3 and DeepSWE hit 77.0 vs Opus 5's 74.0, though HLE and terminal tasks still trail; outputs are shorter and faster, with a notable jump in 3D generation.

A user reports Codex just upgraded DSH from 0.1.1-rc.1 and is running a small live request test against DeepSeek-V4.1-Flash 1017.

Collects the launch thread, release notes, MIT-licensed HF weights, ModelScope mirror and Search & Fetch docs, noting all scores are DeepSeek's own and Opus 5 still leads on HLE, Terminal-Bench and others.

Release note: the model is available on DeepSeek's site, app and API, and users should upgrade DeepSeek Harness to v0.2.1-rc.1 or later.

A user plans to run V4.1 Flash FP8 locally on two Mac Studio M5 Ultra 512GB machines, expecting a solid 80 tok/s.