Modal

AI infrastructure that developers love 💚

Run inference, sandboxes, batch processing, training, and many other things on Modal

[Modal @modal]

Today we're announcing that Modal has crossed $1B in annualized run rate, as well as raised a $750M Series D at a $15.75B valuation, led by @Accel.

Since our Series C, usage has tripled and we’ve grown more than 15x year-over-year.

We are deeply grateful to the developers and companies building on Modal. There's still so much more to build, and we can't think of a better time to be building software.

[Erik Bernhardsson @bernhardsson]

https://t.co/nByTwj2JPX

[Modal @modal]

Read more at https://t.co/lNvBGBIOxn

[Modal @modal]

Performant reinforcement learning is a hard infrastructure and research problem. We’ve recently made on-demand multi-node clusters available and showed how to take advantageous of heterogeneous sampling compute. But there still exists the need for high-fidelity environments, clear and robust reward functions, and tuned training recipes that actually work.

Most teams are actually nearly there! With an agent harness deployed in production and evals/agent traces collected from that deployment, you just need an easy way to start training your data. So we built a training library that gives you the building blocks, tuned defaults, and observability you need for easy, production-grade LLM training.

It’s available now. All it takes is a prompt.

[Modal @modal]

‍@AfterQuery curates long horizon data from domain experts for models to be trained on. Using our training library, they were able to kick off autoresearch for models like Kimi and DeepSeek at 256k context length on their existing datasets.

[Datadog, Inc. @datadoghq]

✅ 96% of issues resolved before customer impact.
✅ 99% of incidents detected within five minutes.

Read the full story to see how the team at @modal use Datadog to operate AI infrastructure across 20+ cloud providers, launching 26M containers every day while maintaining unified visibility across the platform: https://t.co/uMVuRWmlEz

[Daniel Shaar @dshaar_]

https://t.co/qp9PMs5DXv

[Modal @modal]

Voice agents, multiplayer games, and other workloads need to maintain continuity between requests.

With Sticky Sessions, built on Modal's serverless Servers, you can route requests in a session to the same container and autoscale with demand.

Here's a look at how we built it:

[Daniel Shaar @dshaar_]

https://t.co/qp9PMs5DXv

[Modal @modal]

Congrats to @GradiumAI on launching Phonon!

We built a voice agent with their new TTS models and Pipecat on Modal to put them to work.

Read more about Phonon and check out the demo: https://t.co/BkC560ZONl

[Gradium @GradiumAI]

A new frontier for on-device speech.

We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.

Phonon-31m is available under Apache 2.0.

https://t.co/d3qa9qmnzY🧵

[Charles 🎉 Frye @charles_irl]

@modal and @GrayAreaorg are partnering to put the "art" back in "artificial intelligence".

We've sponsored a dozen projects by leading digital artists that will be shown at Gray Area next week.

Come thru on Thursday for a panel featuring @ideeplistening and @Ryphi10.

Register here: https://t.co/FW8rll2A9V

[Modal @modal]

More than 40x the parameters of Mercury 1, still running at 1,000+ tokens/sec.

Congrats to the @_inception_ai team on Mercury 3!

[Stefano Ermon @StefanoErmon]

1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.

[Modal @modal]

Read about how we benchmarked Vera, what we found, and why we're excited about its potential for Sandboxes: https://t.co/zgLJWb1K7W

[Modal @modal]

We got our hands on @nvidia's new Vera CPU and put it to the test.

We built a benchmark using real-world agentic workloads to see how it performs under contention.

In early tests, Vera delivered up to 2x the throughput of comparable x86 hardware, maintaining low latency even as hundreds of concurrent workloads competed for resources.

For Modal Sandboxes, that could mean running more agents in parallel without sacrificing latency.

[Modal @modal]

When we introduced VM Sandboxes, we wanted to preserve Modal’s sub-second cold starts without compromising performance.

We ended up building a new filesystem stack that dramatically improved performance, and wrote about the engineering challenges and decisions that got us there.

[Amit @amitpr]

The secret to Modal's speed is our custom lazily loaded container filesystem, ImageFS. ImageFS is built on a massive tiered distributed filesystem, layering local NVMe disks, regional and p2p caches, and global blob stores.

When we launched VM Sandboxes this month, we wanted to make sure they could take advantage of ImageFS in the same way our gVisor containers do. To do that, we built out a new filesystem stack that tricks a VM's ext4 driver into thinking it's talking to a normal disk.

[Modal @modal]

@ericyuegu @ajhinh 💚🚀

[Eric Gu @ericyuegu]

1/ I trained an AI to be top 1% in the world at Super Smash Bros. Melee by watching 80k hours of human replay.

Meet HAL.

[Modal @modal]

Learn more and deploy Reflection Beam as a Dedicated Endpoint on Modal: https://t.co/idMbhOfjeQ

[Modal @modal]

Reflection Beam is now available on Modal.

Beam is a 501B-parameter open-weight MoE (23B active) from @reflection_ai built for coding, agentic workflows, and reasoning. It supports a 1M token context window, with token efficiency a core focus.

In early access, the Oura Health Intelligence team found that "Beam had fewer context leaks and was much more grounded in its answers than other frontier models we evaluated.”

[kwindla @kwindla]

Fast is better than slow.

We've got two new PhoneLLM models optimized for conversational voice AI.

PhoneLLM Mini (30B) — When you need the lowest possible latency (sub-100ms TTFAT), smallest model size, and lowest cost. Good multi-turn conversation behavior and tool-calling accuracy in a model that can run >100 concurrent agents on a B200 and fits on an RTX 5090.

PhoneLLM Max (120B) — A "microthinking" model for complex conversational tasks. Sub-200ms TTFAT. SOTA task completion performance at 1/10th the latency and cost of models with similar capabilities.

Both of these models are full-weights fine-tunes of @NVIDIAAI Nemotron models. Mini is based on Nemotron 3 Nano. Max is based on the brand new Nemotron 3.5 Super VL.

Both of these models are available as open weights on HuggingFace, and from four inference partners: AWS, Baseten, Cloudflare, and Modal.

[Modal @modal]

Learn more and deploy Nemotron 3.5 Super VL as a Dedicated Endpoint on Modal:
https://t.co/lBYyP9QMh6

[Modal @modal]

Nemotron 3.5 Super VL from @nvidia is live on Modal today with 120B params for visual reasoning and agentic work across images, docs, and UIs.

Hear more about the thinking behind the Nemotron family: At Runtime, @ctnzr explained why “efficiency is also intelligence,” and how NVIDIA pre-trained Nemotron 3 Ultra entirely in FP4.

[Modal @modal]

Watch the full talk: https://t.co/vTb2QYYi5E