Purism has announced the Librem 5 v2, and it's a big change: instead of another ARM chip from NXP (or another vendor), the new phone uses the SpacemiT K1, an eight-core RISC-V SoC. Since we've finally got a successor to the Librem 5 on the table, I wanted to check how the K1 compares to the i.MX 8M Quad in the current Librem 5, have a short look back at the original crowdfunder, and offer my personal opinion.

What's been announced

According to the campaign page, the Librem 5 v2 brings:

  • SoC: SpacemiT K1, 8× X60 RISC-V cores (vs. 4× Cortex-A53 in the i.MX 8M Quad)
  • RAM: 8 GB (vs. 3 GB), "at least" for initial production
  • Storage: 128 GB eMMC, plus a faster SD slot (SDR104, ~80 MB/s)
  • Display: 5.7″, 720×1440, same as before
  • Cameras: 13 MP rear, 8 MP front, with HD video and hardware encoding
  • Connectivity: a new 4G modem, dual SIM, the familiar hardware kill switches
  • Boot: U-Boot, like the original

It's a crowdfunding campaign again. The campaign runs until February 15, 2027. The DevKit is expected 6–7 months after that, and the phone 14–15 months after it, so think spring 2028 at the earliest. The phone-only reward costs USD 1,599, and Purism itself warns that the final retail price will likely be higher.

Purism doesn't publish any benchmarks, only claims: "cooler, faster, longer battery life".

Looking back: the 2017 campaign

This isn't Purism's first phone crowdfunder, so it's worth remembering how the last one went. The original Librem 5 campaign launched on August 24, 2017, with the phone "starting at $599—less than the cost of many popular smartphones". It passed its USD 1.5 million goal in October 2017. I placed my order on October 23, 2017. The old crowdfunder page is still available on archive.org and may be worth a look.

The pitch was about more than hardware. Purism partnered with Matrix and called the Librem 5 "the first ever Matrix-powered smartphone, natively using end-to-end encrypted decentralized communication in its dialer and messaging app". "Make encrypted calls that mask your phone number" was at the top of the feature list. A few weeks later Purism announced a collaboration with Monero, "integrating Monero into Purism's Librem 5 smartphone as part of its default mobile payment system". Days later came a partnership with Nextcloud, promising "Nextcloud applications built into the Librem 5" for "convenient, ethical encrypted file storage".

Neither the Matrix nor the Monero promise was kept. Chats did eventually learn Matrix messaging, but the dialer is a regular phone dialer, and encrypted calls that hide your number never arrived. Monero never became a default payment system on the phone, which IMHO, as a crypto sceptic, is totally fine. Nextcloud fared a little better: through GNOME Online Accounts you can sync files, contacts and calendars with a Nextcloud server. But that's what any GNOME desktop stack used with Phosh gives you, and it's nowhere near the deep integration the announcement suggested.

Other parts of the vision did work out, and they matter more than any single feature. Purism put real money and developer time into the GNOME stack. Phosh, the mobile shell Purism built for the Librem 5, now ships on postmarketOS, Mobian and other distributions, on phones Purism never made. libhandy, which Purism started to make GTK apps adapt to small screens, grew into libadwaita, the library most new GNOME apps are built on. That's a big part of why so many Linux apps work on a phone today. The partnership with KDE didn't produce much, but thanks to distributions like Mobian and Nura (formerly postmarketOS), you can install and use Plasma Mobile on the Librem 5. Purism made a lasting contribution to mobile Linux, and I'm grateful for it.

Then there were the delays. The campaign targeted January 2019 for shipping. Instead, dev kits and early batches trickled out from late 2019, mass production only started at the end of 2020, and COVID-19 and the chip shortage hit. My October 2017 order arrived in March 2022, almost four and a half years after I placed it.

The delays led to the biggest controversy of the campaign. In 2020, Purism changed its refund policy: people who cancelled their Librem 5 order would only get their money back once the phone had shipped. Years later, many of them are still waiting. I never felt like the Librem 5 was a scam, as I did get my phone in the end, but I understand why some people felt it was one. As Amos Batto put it in his recent look at Linux phones: "Refunding the 600 customers who are still waiting for their Librem 5 refunds would do a lot to restore the company’s public reputation, and it would cost the company less than $400k." The number comes from a 2023 letter from Purism to Louis Rossmann, so it may have changed since.

The v2 campaign leaves no room for that kind of dispute. Its FAQ answers "Can I get a refund?" with "No. There are no refunds.", and the list of risks adds that "rewards are not guaranteed to be delivered". That's clear and honest, but it's also bleak: once you've paid, your money is gone, whether or not the crowdfunding goal is met or a phone ever arrives.

Now compare the prices: USD 599 then, USD 1,599 for the phone-only reward now. Even adjusted for inflation, USD 599 in 2017 is roughly USD 800 today. And if the v2 timeline holds, the phone will arrive more than ten years after the first campaign launched. Purism is asking backers to pay twice as much, with no way out, for a schedule that starts with similar unknowns. Given the last campaign, I'd take that timeline with a grain of salt. On the plus side, many parts seem to stay the same. It very much looks like they're keeping the case (maybe with a different SIM tray, unless the second SIM is an eSIM), and likely components like the cameras. Fewer changes should mean less investment and less risk.

CPU performance: don't expect a leap

I found no direct comparison, so here are public Geekbench uploads. The Geekbench 6 Librem 5 result is from my own phone running PureOS Crimson. The others come from different boards and OS builds, so treat them as rough.

Single-coreMulti-core
Geekbench 5
i.MX 8M Quad, 4× A53 @ 1.5 GHz (1, 2)139510
SpacemiT K1, 8× X60 @ 1.6 GHz (link)84497
Geekbench 6
Librem 5 (r4), i.MX 8M Quad, 4× A53 @ 1.5 GHz (link)141452
i.MX 8M Mini, 4× A53 @ 1.8 GHz, for reference (link)156437
SpacemiT K1, 8× X60 @ 1.6 GHz (link)134583

The K1's Geekbench 5 single-core score is dragged down by very weak crypto and floating-point sub-scores, partly because that version has no optimized RISC-V paths. Geekbench 6 is fairer, and here I can compare against an actual Librem 5 run. Single-core, the K1 lands about 5% below the Librem 5 (134 vs. 141), so one X60 core is roughly a Cortex-A53. Multi-core, its eight cores give it about 29% more throughput (583 vs. 452).1 The sub-scores vary a lot. The A53 is more than twice as fast in the single-core Ray Tracer test (175 vs. 77), while the K1's vector unit gives it a big lead in Object Detection and Background Blur.

So the CPU won't get much faster. Day to day, the bigger wins should come from the other changes: 8 GB instead of 3 GB of RAM, much better memory bandwidth, faster storage, and hardware video encoding.

GPU: from etnaviv to "being worked on"

The i.MX 8M Quad has a Vivante GC7000Lite. It's not fast, but it's covered by etnaviv in mainline Mesa and the kernel, and that driver has been in daily use on the Librem 5 for years. OpenGL ES 3 support landed recently, which matters, as UI frameworks have been dropping OpenGL ES 2 support.

The K1 has an Imagination PowerVR BXE-2-32. SpacemiT's own Bianbu OS uses Imagination's closed driver stack, shipped as binary .so libraries. On the open side, Imagination now has an upstream kernel driver (drm/imagination) and a Vulkan driver in Mesa, plus OpenGL (ES) via Zink. But Mesa's PowerVR documentation still lists the BXE-2-32 as unsupported and not under active development. Imagination has started releasing firmware for it, which is a good sign, and Purism budgets a six-month "graphics acceleration" effort. Still, a working open GPU stack on this chip is an expectation for now, not a fact.

Mainline support: good for a headless board, not for a phone

To be fair, the K1 has come a long way upstream since initial support was submitted in mid-2024. Mainline now has Ethernet, SD/eMMC, PCIe, USB 3, SPI, cpufreq and thermal support, and SpacemiT is actively upstreaming. That's enough for a headless single-board computer.

A phone needs much more. Above all, there's no upstream display driver for the K1 yet. Patches for a SpacemiT DRM driver are under review, but they target the newer K3. The K1's vendor kernel uses a display block from the same "Saturn" family, so some of that work may carry over eventually. Then there's MIPI DSI, cameras, audio, power management for suspend, and the modem path.

Purism's campaign page says the same thing: most peripherals aren't in mainline, and drivers will be "adapted and ported" from SpacemiT's 6.6 vendor kernel. For a company whose main selling point on the Librem 5 has been its upstream work, the v2 basically starts that process over.

The new modem

While the BroadMobi modems used in the Librem 5 won't be missed, there are two things worth noting here. First, dual SIM support will require work in ModemManager. To the best of my knowledge, it supports neither dual-active nor multi-standby operation, and there's no automatic failover: ModemManager only switches SIMs when told to. Second, there's no 5G. Adding it may require work on the antennas and the case, at least that's how it has been discussed on the Purism forums.2

Cost

SpacemiT doesn't seem to publish K1 chip prices, but boards give a hint. Orange Pi sells a complete K1-based router board for USD 30, RAM and eMMC included. Meanwhile, the i.MX 8M Quad still sells at USD 50–75 for single units and around USD 29 at 200+ (and is marked "not recommended for new designs"). The K1 is clearly the cheaper chip, by a wide margin. But it's a small part of a USD 1,599 phone, and SoC cost obviously isn't what drives that price.

My take

Leaving NXP's ageing i.MX 8M makes perfect sense, and the K1 is cheap, available, and gets some amount of upstream attention. But on paper it doesn't make the phone much faster. I thus doubt the long-term usefulness of the device, as I don't expect even RAMageddon to make websites lighter going forward. Its GPU has no supported open driver yet, and most of what a phone needs isn't in mainline. On top of that, it's an RVA22 chip at a time when the RISC-V software ecosystem is moving to RVA23,3 and that will hurt software availability. There's no Flathub.org support for RISC-V yet, and if it ever lands, it may well require RVA23. So which apps users get will depend on Purism putting in the work to package the relevant mobile Linux apps, in Debian or in a Flatpak remote of their own. IMHO, aside from lofty claims, the only benefit of choosing RISC-V over a newer ARM chip is that there are few (if any) RISC-V phones, which makes it easier for Purism to claim that what they do is unique.

Purism had to put in years of upstreaming work before the original Librem 5 was in good shape, and the v2 will need the same. I'll be watching how they tackle it, but only from the sidelines: I won't pay USD 1,599 for it.4 Unless you have money to burn or are a die-hard RISC-V enthusiast, I advise you to sit this one out, too.

1

The Geekbench 6 Librem 5 result is a single run (Geekbench 6.7.1 on PureOS 11 Crimson, uploaded November 14, 2026), and the K1 run used a different Geekbench 6 build on Bianbu, so small differences are within noise. I've kept the i.MX 8M Mini, which has the same Cortex-A53 cores clocked at 1.8 GHz, as a sanity check. The two i.MX 8M Quad Geekbench 5 links come from the same uploader: the first is labelled "i.MX8MQ, 4x Cortex-A53 @ 1.5GHz", the second has the scores. If you own a Librem 5 and have run Geekbench 6 on it, please send me your results too!

2

I failed to find the actual post I am vaguely remembering, but this thread and this other thread about using a 5G modem in the Librem 5 are interesting reads.

3

I've had a VisionFive 2 for a while now, and it has taught me how messy RISC-V gets when every chip implements a slightly different set of extensions. That's why I agree with Canonical that RVA23 is a sensible baseline for RISC-V: it makes things like the vector and hypervisor extensions mandatory, so distributions can actually build for them. Ubuntu requires RVA23 since 25.10, and 26.04 LTS keeps that baseline. The K1 is an RVA22 chip with RVV 1.0, so it isn't RVA23, which means it stays on Ubuntu 24.04 LTS. PureOS is Debian-based and Debian hasn't made that move, so this won't break the Librem 5 v2 tomorrow. But a phone that ships in 2028 on a pre-RVA23 chip is not betting on where RISC-V is going.

4

I might have for a Librem 5 v2 with a decent ARM chip. Even a modest one like the Rockchip RK3566 would have made me more positive, because going RISC-V adds a lot of risk through unclear software availability. That said, if an upgrade board ever becomes available, I may order one for my Librem 5.