# DeepSeek V4 Flash 828 Hacker News commenters recommended DeepSeek V4 Flash and 223 criticized it, November 2025 to November 2026. It was recommended in 8 of 13 months; owners who said how long report a median of 0.04 years (24 owners). It meets the Hacker Approved badge rule. **Verdict:** A very cheap, fast model that people find close to frontier quality for coding and everyday tasks; critics say it errs more, loops, and needs steering. ## Why people recommend it - Extremely cheap to use (13 comments): "My personal favorite is DeepSeek v4.1 Flash but I have tried Qwen 3.8, Kimi 3 and GLM 5.3 which are equally impressive but DeepSeek is the cheapest and fastest" ([source](https://news.ycombinator.com/item?id=49897132)) "I was running deepseek v4.1 pretty much non stop during work hours, with heavy tool/mcp usage and finding it very difficult to spend more than $75 in a month." ([source](https://news.ycombinator.com/item?id=50013207)) "I have been loving deepseek flash. I can code in my hitl preference doing complex but moderate amount of coding and not even hit 1 dollar in a session." ([source](https://news.ycombinator.com/item?id=49124392)) - Near-frontier coding quality (9 comments): "it has proven so capable that after being simultaneously on Max x20 and Pro x20 subscriptions, i've transitioned to using DS 4.1 as a daily driver and am very satisfied" ([source](https://news.ycombinator.com/item?id=49905220)) "I dont even bother asking opus those kind of questions when I know deepseek-4.1-flash or glm-5.3* or mistral-large-4 will give the proper solution" ([source](https://news.ycombinator.com/item?id=50065703)) "one-shotted by deepseek-v4.1-flsh-1017 (in high mode, not even max) with sound and programmatically generated assets in under 5 minutes" ([source](https://news.ycombinator.com/item?id=50133319)) - Open weights, runs locally (5 comments): "We can run several coding agents during the day and batch inference jobs overnight and serve the entire team with guaranteed privacy" ([source](https://news.ycombinator.com/item?id=49170504)) "play with models that are cheap but bigger than what you'd reasonably be able to run locally like deepseek v4 flash (unquantized)" ([source](https://news.ycombinator.com/item?id=48726111)) "DS Flash is where local models begin to feel useful for coding, but the quants we run locally are sharply reduced in intelligence" ([source](https://news.ycombinator.com/item?id=49155775)) - Very fast (3 comments): "My personal favorite is DeepSeek v4.1 Flash but I have tried Qwen 3.8, Kimi 3 and GLM 5.3 which are equally impressive but DeepSeek is the cheapest and fastest" ([source](https://news.ycombinator.com/item?id=49897132)) "I swear to god that DeepSeek V4-Flash is the most useful model available right now. It's SO FAST and is good enough for so many tasks" ([source](https://news.ycombinator.com/item?id=48191894)) "tiny, highly cached, fast models like Qwen's 27b/a3b range or DeepSeek flash are highly capable and relatively inexpensive to run" ([source](https://news.ycombinator.com/item?id=49426636)) ## What people warn about - Mistakes and bad tool use (5 comments): "V4 Flash 7/31 suffers from a lot of pathological behavior in coding harnesses as well, e.g. infinite loops, hallucinations, premature termination, and invalid tool calls." ([source](https://news.ycombinator.com/item?id=49626609)) "The flash version is great at implementing a plan but because it makes trade-offs for increased speed it doesn't take the time to think through everything" ([source](https://news.ycombinator.com/item?id=50027699)) "deepseek-v4-flash tends to ignore instructions to use these tools and default to churning through code and guessing the cause of errors" ([source](https://news.ycombinator.com/item?id=48463687)) - Gets stuck in endless loops (3 comments): "V4 Flash 7/31 suffers from a lot of pathological behavior in coding harnesses as well, e.g. infinite loops, hallucinations, premature termination, and invalid tool calls." ([source](https://news.ycombinator.com/item?id=49626609)) "I've definitely noticed that a high level of deepseek flash responses were looping endlessly before the 0731 release." ([source](https://news.ycombinator.com/item?id=49665208)) "I've been using Q4_K_S for a while now and it's been rock solid, but having tried Q3_XS, that would loop like crazy." ([source](https://news.ycombinator.com/item?id=50133423)) - Too big for local use (3 comments): "DS Flash is where local models begin to feel useful for coding, but the quants we run locally are sharply reduced in intelligence" ([source](https://news.ycombinator.com/item?id=49155775)) "The bad news is that the original v4 flash was 284B, which was large but still somewhat reasonable for running locally." ([source](https://news.ycombinator.com/item?id=49639317)) "The DS 0731 was so atrocious it was getting beaten by small Nemotron in my coding tests" ([source](https://news.ycombinator.com/item?id=50304441)) - Needs more steering (2 comments): "I'm fine with DS Flash V4.1 in code bases I have experience in, don't really notice a huge difference other than more steering is needed." ([source](https://news.ycombinator.com/item?id=50092943)) "Rust works perfectly fine, but when I use Rust, I usually pay closer attention to performance, so I have to guide it a bit" ([source](https://news.ycombinator.com/item?id=49121525)) ## Alternatives hackers recommend - [Claude Opus](https://hackerapproved.com/p/claude-opus.md): 3331 recommend - [OpenAI GPT](https://hackerapproved.com/p/gpt-5-5.md): 1205 recommend - [Claude Sonnet](https://hackerapproved.com/p/claude-sonnet.md): 854 recommend - [Fable](https://hackerapproved.com/p/fable.md): 823 recommend - [Claude Fable 5](https://hackerapproved.com/p/fable-5.md): 774 recommend ## Where to get it - [Visit the official site](https://www.deepseek.com/) ## Cite as Hacker Approved, "DeepSeek V4 Flash", https://hackerapproved.com/p/deepseek-v4-flash/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).