DeepSeek V4.1 Flash hands-on tests and comparisons

On the official API with thinking on, fixing json_schema to json_object raised scores to 92/100 at temp 1.0 and 90/100 at temp 0, up from 87 and 85.

DeepSeek says the new model scores higher, runs faster and outputs less, with MIT-licensed weights on Hugging Face and Search/Fetch APIs open to all at $0.75 per 1,000 calls.

Usage feedback: a user wants DeepSeek-V4.1-Flash-1017 available for free inside OpenCode.

Benchmarks show Terminal-Bench 4.0 up from 31.2 to 42.3, DeepSWE 74.2 to 77.0 and Automation-Bench 54.8 to 60.5, with shorter outputs, lower cost and MIT-licensed weights on Hugging Face.

A post argues DeepSeek V4.1 Flash codes at roughly Opus 5 level for about 2% of the cost at 200+ tps, fueling a pricing race among labs.

DeepSeek opened its search and fetch API free through Oct 31 for Claude Code, Codex and DeepSeek Harness (normally $0.75 per 1,000 searches), while V4.1 Flash tokens still bill from $0.15 per 1M input.

Official announcement cites benchmark gains, faster task completion and shorter output time; live on web, app and API, with Harness v0.2.1-rc.1+ required.

The post says OpenCode, a terminal coding assistant with selectable models, currently offers DeepSeek V4.1 Flash for free.

Claims broad coding and agent gains, DeepSWE 77.0, faster and cheaper output at unchanged price with open weights, plus a domestic and international positioning take.

A commentary argues the real news is DeepSeek's first-party Search and Fetch APIs at $0.75 per 1k calls on the same key, free in Harness/Claude Code/Codex until Oct 31, threatening middleware like Tavily, Exa and Firecrawl.

Summarizes DeepSeek's own numbers: 77.0 on DeepSWE v1.1 versus 74.2 before, 74.0 for Opus 5 and 73.0 for Sol, while Opus 5 still leads HLE and Terminal-Bench; MIT weights, unchanged API pricing, Search/Fetch at $0.75 per 1,000 calls.

Web and app updated with API still on deepseek-flash at unchanged price; lists HLE 42.6, Terminal-Bench 4.0 42.3, DeepSWE v1.1 77.0 and says runtime and output length are shorter.