Is Deepseek V4 really out?
The question surfaced on April 25th, and it was almost disarmingly simple: is Deepseek V4 really out? No press release, no model card, no post from Deepseek's own accounts — just a community asking itself whether something had already happened while it wasn't looking. That framing matters. Deepseek has a documented habit of pushing models into the world quietly, letting the benchmarks and the forum posts do the announcing. That history is what gives the murmur traction. V3 appeared with relatively little ceremony before becoming one of the more discussed open-weight releases of recent memory. So when someone asks whether V4 has slipped out in the same fashion, the question doesn't sound absurd. It sounds like a pattern. As of now, nothing has emerged to either confirm a release or definitively close the door on one — no weights, no confirmed API updates, no corroborating signals from researchers who would typically be among the first to notice.
If confirmed, here is what this means. A Deepseek V4 release — even a quiet one — would arrive at a moment when the open-weight model landscape is under more scrutiny than ever. V3 already forced a recalibration of expectations around what non-frontier labs could produce, and another generation from the same team would sharpen that pressure considerably. Western AI companies watching their closed models lose ground to Chinese open-weight releases would face an uncomfortable conversation about pricing, accessibility, and the durability of their moats. Developers building on top of current models would have immediate, practical decisions to make. The strategic signal — that Deepseek is continuing to iterate at pace — would matter as much as the technical benchmarks.
Watch for any corroborating activity: new model weights appearing on Hugging Face, changes to Deepseek's API documentation, or benchmark results surfacing from researchers who would plausibly have early access. An official announcement, if one comes, would likely follow that trail rather than lead it.
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1su3fle/deepseek_v4_flash_pro_now_out_on_api/"> <img alt="DeepSeek V4 Flash & Pro Now out on API"
<!-- SC_OFF --><div class="md"><p>Hey 👋</p> <p>Saw the tweet making the rounds about deepseek v4 being 35x cheaper than opus on input and 178x cheaper on cached tokens, and was
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1su5gj5/buried_lede_deepseek_v4_flash_is_incredibly/"> <img alt="Buried lede: Deepseek v4 Flash is
<!-- SC_OFF --><div class="md"><p>DeepSeek-V4.1-Flash's prompt state is only 0.9 KB/token, so you can split it at layer 20 across CUDA/Metal. All you need is 1/10GbE network.
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1su3wf7/deepseek_v4_released/"> <img alt="Deepseek V4 Released"
<!-- SC_OFF --><div class="md"><p>The benchmark sheet dropped this morning and people are losing it in the ML community.</p> <p><strong>What DeepSeek R2 scores:</strong><br />
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1x8qvaj/muse_spark_on_m3u_256_gb/"> <img alt="Muse Spark on M3U 256 GB"
<!-- SC_OFF --><div class="md"><p>so v4 pro dropped and barely anyone is talking about it. feels weird since when kimi k2.6 came out i seen post about it everywhere</p> <p>anyone
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →