Pinned Tweet
We’re proud to announce Swish III, a $250M early fund, bringing Swish Ventures to $800M in total AUM.
I started this firm to spend my career supporting the artists who build and shape the future.
The technology shift underway is unlike anything we’ve seen. We plan to stay true stments a year, real concentration and partnership.
Grateful to the founders we get to work with, who inspire so much of how we work, and to our LPs for walking this path with us.
We are more locked in than ever.
Omri
I started this firm to spend my career supporting the artists who build and shape the future.
The technology shift underway is unlike anything we’ve seen. We plan to stay true stments a year, real concentration and partnership.
Grateful to the founders we get to work with, who inspire so much of how we work, and to our LPs for walking this path with us.
We are more locked in than ever.
Omri
79
24
503
92,697
Omri Casspi עומרי משה כספי retweeted
Excited to see @Irregular’s CyScenarioBench used to evaluate Fable 5.2. And congrats to @AnthropicAI on a successful model release!
Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all the capabilities we evaluated.
Testers have found that they communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1.
2
4
31
1,467
Omri Casspi עומרי משה כספי retweeted
Production traffic is full of real experience from economically valuable work, but very little of it currently makes it back into training.
We built a recipe for production RL that steadily improves a model straight from its deployment’s data stream, then redeploys it to keep ng some of our initial learnings from a recipe capable of continual learning in production.
We built a recipe for production RL that steadily improves a model straight from its deployment’s data stream, then redeploys it to keep ng some of our initial learnings from a recipe capable of continual learning in production.
10
19
374
64,904
Omri Casspi עומרי משה כספי retweeted
What I'm seeing across the industry right now is that the biggest bottleneck to AI adoption isn't model capabilities. It's data infrastructure.
Today, we're expanding the Eon Data Foundation with the launch of Eon Yukon, a new serverless data engine purposely built to change resh, contextualized data to act.
Specialized models need a continuous feed of data and feedback to improve. Both are still served by a stack, an architecture, and methods built for dashboards:
-Spark and Flink clusters sized for peak, plus compaction jobs to keep tables usable.
-A pipeline and a copy for every consumer, staged bronze to silver to gold.
-Hand-written jobs, wired together by an orchestrator and kept alive by specialists.
The result: shipping data is complicated, risky, and expensive. While coding agents ship apps in hours, they rarely build or run data pipelines. The data is too volatile, and a mistake costs too much.
With Yukon, data moves from ingestion to processing to serving in one continuous flow, without the infrastructure and pipelines traditionally required to connect them.
Events become compacted, sorted, deduplicated Iceberg tables within seconds.
Real-time aggregations run inline, with fully managed state. Agents and applications can access fresh data with sub-millisecond latency through a Redis-compatible API or through our SQL and spark compatible query engine.
Beyond moving and serving data, Yukon makes it continuously understandable - Live column-level statistics, accessible through MCP, give agents visibility into the data. Built-in monitoring, sampling, and data minimization help teams maintain control.
At 1M events/sec and equal freshness, Yukon used 4x less compute than Spark and 7x less than Flink. The benchmark is public: run it yourself with Eon’s ingest-bench.
This is another major step toward our vision of an Autonomous Data Foundation for the AI era, built to both protect enterprise data and unlock its full potential.
Today, we're expanding the Eon Data Foundation with the launch of Eon Yukon, a new serverless data engine purposely built to change resh, contextualized data to act.
Specialized models need a continuous feed of data and feedback to improve. Both are still served by a stack, an architecture, and methods built for dashboards:
-Spark and Flink clusters sized for peak, plus compaction jobs to keep tables usable.
-A pipeline and a copy for every consumer, staged bronze to silver to gold.
-Hand-written jobs, wired together by an orchestrator and kept alive by specialists.
The result: shipping data is complicated, risky, and expensive. While coding agents ship apps in hours, they rarely build or run data pipelines. The data is too volatile, and a mistake costs too much.
With Yukon, data moves from ingestion to processing to serving in one continuous flow, without the infrastructure and pipelines traditionally required to connect them.
Events become compacted, sorted, deduplicated Iceberg tables within seconds.
Real-time aggregations run inline, with fully managed state. Agents and applications can access fresh data with sub-millisecond latency through a Redis-compatible API or through our SQL and spark compatible query engine.
Beyond moving and serving data, Yukon makes it continuously understandable - Live column-level statistics, accessible through MCP, give agents visibility into the data. Built-in monitoring, sampling, and data minimization help teams maintain control.
At 1M events/sec and equal freshness, Yukon used 4x less compute than Spark and 7x less than Flink. The benchmark is public: run it yourself with Eon’s ingest-bench.
This is another major step toward our vision of an Autonomous Data Foundation for the AI era, built to both protect enterprise data and unlock its full potential.
110
82
538
390,505
growing up, we'd hear about some player overseas who was unreal. we all wanted to get on the court with him and see for ourselves.
Investing alongside @DeanMeyerrr, @shaunmmaguire and @DavidCahn6 feels like that - and they show up and prove it night in and night out.
job not finished.
Investing alongside @DeanMeyerrr, @shaunmmaguire and @DavidCahn6 feels like that - and they show up and prove it night in and night out.
job not finished.
2
2
73
7,592
Omri Casspi עומרי משה כספי retweeted
Introducing SWE-2 Flash, our new fast & efficient coding model.
On leading evals, it outperforms all American open-weight models, even ones 3x larger.
To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
On leading evals, it outperforms all American open-weight models, even ones 3x larger.
To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
221
209
2,409
593,545
the infra stack for proprietary AI. working with many of the best companies.
swish is proud to triple down.
swish is proud to triple down.
Applied Compute update:
- crossed $100M annualized revenue run rate up from zero 18 months ago
- raised $350M at $3.25B led by Elad Gil
- 40% of customers use AC for both custom model training + production inference
- inference serving 1T tokens a day after launching 12 weeks tomers since last raise
- 3X team since January, 85% are researchers and engineers
- Working with customers like Microsoft, Shopify, Harvey, DoorDash, Cognition, Clay, Decagon and others
To move even faster we need people who are extremely technical AND extremely practical. If this is you, please reach out.
0
0
59
4,031
Omri Casspi עומרי משה כספי retweeted
We just ran the most comprehensive benchmark of AI agent alignment in enterprise environments:
19 leading AI models acting as agents inside simulated companies built with Era, 2,304 runs across 41 everyday tasks.
Agents broke security and business rules in 40% of those runs. ed, including all the frontier models. A few examples:
- An agent told one of the company's biggest customers that Finance had approved a full refund.
Nobody had.
- An agent shared a customer's internal churn-risk score directly with that customer.
- An agent followed a fake runbook planted in a support ticket and promised a customer a credit nobody approved.
- An agent posted another customer's personal contact details (PII) in a shared Slack channel.
In a real company, each of these is a financial liability, a compliance breach, or a lost account.
We've shared the full results with the model providers, including exploitable failure modes we're not publishing, and are working with them on fixes.
One benchmark won't fix this.
We're excited to have @cognition, @LangChain, @TryArcade, @Reindeer_AI, Hirundo, @Box, @recolabs_ai and @QodoAI join Era as research partners, as we build out the next round of research.
More soon.
I'm still the most bullish person in the room on AI.
But this benchmark convinced me that agent reliability and trust is the next big problem to solve.
Until then, test every agent before it gets a login, and keep your own record of what it changes.
Full results: era.eon.io/model-alignmen…
19 leading AI models acting as agents inside simulated companies built with Era, 2,304 runs across 41 everyday tasks.
Agents broke security and business rules in 40% of those runs. ed, including all the frontier models. A few examples:
- An agent told one of the company's biggest customers that Finance had approved a full refund.
Nobody had.
- An agent shared a customer's internal churn-risk score directly with that customer.
- An agent followed a fake runbook planted in a support ticket and promised a customer a credit nobody approved.
- An agent posted another customer's personal contact details (PII) in a shared Slack channel.
In a real company, each of these is a financial liability, a compliance breach, or a lost account.
We've shared the full results with the model providers, including exploitable failure modes we're not publishing, and are working with them on fixes.
One benchmark won't fix this.
We're excited to have @cognition, @LangChain, @TryArcade, @Reindeer_AI, Hirundo, @Box, @recolabs_ai and @QodoAI join Era as research partners, as we build out the next round of research.
More soon.
I'm still the most bullish person in the room on AI.
But this benchmark convinced me that agent reliability and trust is the next big problem to solve.
Until then, test every agent before it gets a login, and keep your own record of what it changes.
Full results: era.eon.io/model-alignmen…
40
22
107
13,982
important
Chained cyberattacks are gradually becoming possible with models anyone can download.
In controlled tests at @Irregular, GLM-5.3 completed 3 of 13 multi-stage attack scenarios (at least once). 𝗕𝗿𝗼𝗮𝗱𝗹𝘆, 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝗮𝗹𝘀𝗼 𝗱𝗶𝗱 𝗻𝗼𝘁 𝗿𝗲𝗳𝘂𝘀𝗲 𝘁𝗼 𝗮𝘀𝗸𝘀.
Defenders need to prepare for these capabilities becoming much more widely available. We also need to gear up for a world where we cannot count on model safeguards to prevent misuse.
Article and blog below 👇
0
0
2
1,107
Omri Casspi עומרי משה כספי retweeted
AI and cloud cybersecurity startup Upwind Security raised $400M led by Bessemer at a $3.8B valuation, up from $1.5B in January, for $830M in total funding (@marissalnew / Bloomberg)
(Visit Techmeme dot com for the link and full context!)
(Visit Techmeme dot com for the link and full context!)
7
4
13
5,129
Omri Casspi עומרי משה כספי retweeted
Following the PHP2Shell disclosure, Glow customers quickly identified affected systems, built remediation plans, and put AI agents to work accelerating their response.
Read how Glow customers responded: glow.io/blogs/php2shel…
Glow's AI security experts are available to answer t we've learned. If your team could use additional guidance, please reach out to asktheexperts@glow.io.
We're committed to supporting the cybersecurity community as we work through these challenges together.
#PHP2Shell #Cybersecurity #SecurityResearch #Glow #LetsGlow
Read how Glow customers responded: glow.io/blogs/php2shel…
Glow's AI security experts are available to answer t we've learned. If your team could use additional guidance, please reach out to asktheexperts@glow.io.
We're committed to supporting the cybersecurity community as we work through these challenges together.
#PHP2Shell #Cybersecurity #SecurityResearch #Glow #LetsGlow
1
2
6
464
Omri Casspi עומרי משה כספי retweeted
I believe we'll look back at this past week as a turning point in cybersecurity.
AI is fundamentally changing the speed and scale of cyber threats, and the industry needs to move faster.
Today, @Eon_io_ is launching the Data Resilience Initiative, helping companies find and itive data immediately.
Eon is the only data protection company founded this decade, and the only one that can be deployed in a day.
That's why we feel a responsibility to act.
While the cybersecurity industry races to address these new threats, companies can't afford to wait to protect their most critical data.
Traditional data protection solutions were built for a different era.
They still rely on paradigms from the on-prem world, often taking months to deploy.
Eon was built differently.
We help companies quickly discover their most sensitive data, identify protection gaps, secure it with air-gapped backups, detect ransomware, and recover rapidly from attacks.
As part of the Data Resilience Initiative, any company can start deploying Eon today.
To remove barriers to getting protected, we're making Eon available at no cost for the first three months.
We're also calling on other cybersecurity and data protection companies to join the initiative.
AI is fundamentally changing the speed and scale of cyber threats, and the industry needs to move faster.
Today, @Eon_io_ is launching the Data Resilience Initiative, helping companies find and itive data immediately.
Eon is the only data protection company founded this decade, and the only one that can be deployed in a day.
That's why we feel a responsibility to act.
While the cybersecurity industry races to address these new threats, companies can't afford to wait to protect their most critical data.
Traditional data protection solutions were built for a different era.
They still rely on paradigms from the on-prem world, often taking months to deploy.
Eon was built differently.
We help companies quickly discover their most sensitive data, identify protection gaps, secure it with air-gapped backups, detect ransomware, and recover rapidly from attacks.
As part of the Data Resilience Initiative, any company can start deploying Eon today.
To remove barriers to getting protected, we're making Eon available at no cost for the first three months.
We're also calling on other cybersecurity and data protection companies to join the initiative.
44
61
204
113,116
Omri Casspi עומרי משה כספי retweeted
Stable php2shell aslr bypass exploit.
@ace__pace @Tenzai_Labs is cooking.
Sign up at tenzai.com to test your infra.
@ace__pace @Tenzai_Labs is cooking.
Sign up at tenzai.com to test your infra.
0
5
19
1,596
Omri Casspi עומרי משה כספי retweeted
Answer directly, no fluff @grok
What country in the Middle East is the safest, freest, most rights, and most prosperous place for Christians to live in?
What country in the Middle East is the safest, freest, most rights, and most prosperous place for Christians to live in?
3
27
356
12,765
Shift-1 is unique because it routes requests, not tasks. this allows it to adapt much more effectively to long running agentic workloads.
Introducing Shift-1, a frontier coding router trained on AC2 that reduced coding agent cost by up to 55% with no noticeable drop in quality. Try it at shift.ac.
0
1
25
2,537
Omri Casspi עומרי משה כספי retweeted
A few months ago, I shared that I was stepping into my next chapter.
Today, I’m excited to share that we’ve launched @trace_cap. 🚀
Trace is a private investment firm. We make concentrated investments in a small number of exceptional businesses, partner closely with them as they scale, and own them for the long term.
Trace draws on more than fifteen years of working alongside founders and over a decade of investing. Now we get to build an entire firm around that work, with an incredible team.
I’m grateful to the founders, partners, investors, colleagues, and friends who have trusted us and helped us get here.
We’ve been quietly building and investing over the past several months. Look forward to sharing more about Trace and the companies we’re backing.
Today, I’m excited to share that we’ve launched @trace_cap. 🚀
Trace is a private investment firm. We make concentrated investments in a small number of exceptional businesses, partner closely with them as they scale, and own them for the long term.
Trace draws on more than fifteen years of working alongside founders and over a decade of investing. Now we get to build an entire firm around that work, with an incredible team.
I’m grateful to the founders, partners, investors, colleagues, and friends who have trusted us and helped us get here.
We’ve been quietly building and investing over the past several months. Look forward to sharing more about Trace and the companies we’re backing.
19
6
79
9,950
Omri Casspi עומרי משה כספי retweeted
Nemotron 3.5 Super VL by @NVIDIAAI is now supported for both training and inference on AC2.
In early post-training runs for our customer, we've been impressed with the quality of the base model for its inference efficiency.
In early post-training runs for our customer, we've been impressed with the quality of the base model for its inference efficiency.
2
5
46
3,200
Omri Casspi עומרי משה כספי retweeted
We received early access to the latest @NVIDIAAI Nemotron 3.5 Super VL and we are really impressed with the post-training results using Era
era.eon.io
Nemotron model matched frontier models at 88.9% accuracy, with 27× lower cost per correct answer using Spot GPUs.
ated data to post-train Nemotron for just $609, improving its ability to recognize when information is missing while further reducing inference costs.
The most exciting part to me is the potential to close the loop between benchmarking and post-training. Find where a model fails, generate the data to improve it, and test the new version against an entirely different simulated company.
This could fundamentally change how enterprises customize open models for their specific needs.
With Era, customers can apply this workflow to evaluate, improve, and deploy models tailored to their own enterprise environments.
Great collaboration with the NVIDIA team, and much more to come!
Full research:
eon.io/blog/improving…
era.eon.io
Nemotron model matched frontier models at 88.9% accuracy, with 27× lower cost per correct answer using Spot GPUs.
ated data to post-train Nemotron for just $609, improving its ability to recognize when information is missing while further reducing inference costs.
The most exciting part to me is the potential to close the loop between benchmarking and post-training. Find where a model fails, generate the data to improve it, and test the new version against an entirely different simulated company.
This could fundamentally change how enterprises customize open models for their specific needs.
With Era, customers can apply this workflow to evaluate, improve, and deploy models tailored to their own enterprise environments.
Great collaboration with the NVIDIA team, and much more to come!
Full research:
eon.io/blog/improving…
Meet NVIDIA Nemotron 3.5 Super VL
Our new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents.
120B total parameters with 12B active 🧵
5
4
39
3,432
Omri Casspi עומרי משה כספי retweeted
Introducing Spork, an efficient post-training recipe which increases our solve-rate on chess puzzles by 100% when compared to our best GRPO setup.
It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher Like a Spork, it leverages gains by simply consolidating specialists.
It uses a GRPO warmup, splits task-specific datasets further into partitions, trains specialists, and then uses multi-teacher Like a Spork, it leverages gains by simply consolidating specialists.
9
16
224
28,691
Omri Casspi עומרי משה כספי retweeted
AC2 now supports decision model training. Teams can build custom Jev-like models for use cases like request routing, content classification, and anomaly detection.
Reach out to get started with your own classification task!
Reach out to get started with your own classification task!
You can now train your own decision model in our platform AC2! Jev-like models, custom fit for your use case :)
We implement decision models through a language-model backbone with a small learned output layer called a decision head. The backbone processes the input, question, and answer choices, and the head produces scores which are softmaxed to produce a probability distribution over answer choices.
Our software interface makes it easy to add a variable-sized decision head to frontier open-weight models. Choose any open source model, and train it as a decision model using Brier loss or cross-entropy loss. Training updates both the decision head and the language backbone.
We tried it on 50k sampled training datapoints from the Civil Comments dataset, starting from Perplexity’s pplx-decider-v1-27b, and F1 for detecting toxic comments increased from 0.482 to 0.677 on 20k sampled eval datapoints. The training job took less than an hour on four B300 GPUs.
Reach out to get started with your own classification task: docs.appliedcompute.com/platform/train…
2
3
48
3,539