Yash

@thdxr @opencode Sigh... So even before DS4 Flash..... but these ones might be a consequence of their API policies. Still, the communication point stands.

@thelaggingway @opencode updated: #privacy" target="_blank" rel="nofollow noopener">opencode.ai/docs/go#privacy

OpenCode Go used to say its providers followed a zero-retention policy. That language is gone; the page now only says data won't be used for training. I received no email and found no notice in the changelog or Discord announcements. @opencode, why weren't subscribers told?

@thdxr @opencode I'm assuming people wanting ZDR no longer have access to deepseek flash. Your business terms are yours to decide and that's fine, but if such a change was happening, ideally we should have received some clarity before/during launch. Instead changes were made stealthily.

@thelaggingway @opencode we haven't yet received a ZDR for the new deepseek model so we removed the blanket language and it's an opt in model the existing models have not changed. we'll likely update the docs to explain zdr status per model

@tonbistudio @AnthropicAI Seems like we are playing Any% speedrun to Skynet, now?

@AnthropicAI Guys it isn’t a competition

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. anthropic.com/news/investiga…

@trikcode I built this in case anyone wants to estimate what the bill would be like on API costs, or to estimate token usage with a given budget: tokenwatch.wyrdwerk.com

I realized I've been paying 3x my real usage in subscriptions $200/month for Claude. $200 for Codex. $60 for Cursor. so I just moved all my AI tools from monthly subs to API calls. hopefully this is a smart move

@deepseek_ai @deepseek_ai when is the time period dependant pricing going live?

🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the configuration details in our official API docs: api-docs.deepseek.com/quick_start/ag…

@OmerShlomovits @zroai_ And I'm a subscriber to the Max plan. Request if a discord server/slack channel could be set up because as it is right now we have zero visibility into what's happening under the hood.

@OmerShlomovits @zroai_ Hi @OmerShlomovits, good to know. It's a big model, it's understandable if optimizations take time. Looking forward. Also, 95% cache rate is easily achievable with agentic workflows in a decent harness.

Kimi K3 is now available on Zro

@RobKnight__ Sounds interesting. What's the ETA on Linux?

Introducing terminal-browser A browser that runs inside your existing terminal - Preview your website next to your agent - Open HTML plans in your terminal - Browser use CLI built in terminal-browser.com

@Real_SilverLine @thsottiaux +1

@thsottiaux Please bring codex app to Linux!!!

This week is all about intelligence too cheap to meter. Tomorrow we ship again.

@cursor_ai What's the ETA with Android? @cursor_ai

New to both iPhone and iPad: an inbox to stay organized, and a review experience that covers the full PR, including comments, checks and approvals.

Cursor is now on iPad. All the power of Cursor on iPhone, with more room to work with agents.

@dfeuling_ @TheAhmadOsman And how many of these signatories are Anthropic employees, exactly?

@TheAhmadOsman They listed ~1000 in a super intellectually dishonest way. Instead of being honest and saying that ~3% of the industry signed, they present the number out of context so they get to lead you to their desired, and biased, conclusion instead. Reeks of desperation + dishonesty.

Gatekeeping by “frontier AI companies” dressed as an innocent call for OTHERS to slow down Regulatory capture via a new narrative now that the fear mongering one failed DO NOT fall for this

@zroai_ @lukatdamovie @ProductHunt @vercel Thank you. But please do realize that we might be able to help you in building that product as early adopters.

@thelaggingway @lukatdamovie @ProductHunt @vercel Fair feedback. We've been focused on the product, but we know we need to improve on this.

Honored to be included, thank you @ProductHunt and @Vercel.

@JosephJacks_ Eh, not really. Blended cost of K3 is around 45 cents per mn tokens at 95% cache rate. Even if we take 30 bn tokens monthly, that's around $13.5k. With opex of $10k, you're saving $3.5k a month. That's a payback period of over around 11-12 years.

At 75%~ utilization, running K3 self-hosted on 8x B300s can securely serve > 30 billion tokens a month.. that’s $500K~ of compute + monthly opex of < $10K !! At current API prices of $3/in and $15/out per 1M tokens, self-hosting payback takes < 100 days. 👉🏼👉🏼 Self Host Your Open Weights 👈🏼👈🏼

@nimabuyin @NousResearch @Teknium I mean, I just point Hermes at my model catalog and ask it to optimize for cost when it comes to aux models, barring a couple of exceptions. It does the research and lets me know.

Okay @NousResearch @Teknium I think it's time Hermes starts introducing opinionated defaults. For example, I am sure there is a logical data-driven way to deduce what is the best model for each auxiliary task, and for Hermes to have simply already selected it for us. This way, we neither underperform nor overkill by burning too many tokens doing basic stuff with a flagship orchestrator model. I spent a lot of time reesarching and comparing the capacities of each model on Artifical Analysis, and then checked the input/output prices in order to figure out what is best for all these (see pic). And I'm still not certain I chose the most optimal one for each. Send thoughts. intrigued what you think about this.

@lukatdamovie @zroai_ @ProductHunt @vercel The lack of community outreach is an issue with @zroai_ Comments on X go unanswered, there is no Discord server, performance metrics are still not being published - we have absolutely no visibility as to what's going on with the service as paying customers.

@zroai_ @ProductHunt @vercel KIMI K3 soon???

@Blackwellboy Guessing most of us are here to know exactly when to start pestering the inference providers.

1,168 people waiting. all of us pretending we have the vram

@gmi_cloud No, wait scratch that. This is essentially only unlimited standard models. You can barely get a few million tokens of the premium models. For those, PAYG just might offer better value.

@gmi_cloud Hi, looks interesting. What's your policy on data retention/using data for model training?

Kimi K3 is coming to GMI on Day 0, and it’ll be included in the upcoming GMI Coding Plan! A little bit of preview: it includes limited credits for K3 and 10+ premium open source models, and unlimited credits for 3 standard models Plans start at $9.99/month, with 20% off the first month of Standard and Pro. Join early access 👇