CBLab

Our latest cyber-specialized model is out. It's in its own league on many cyber-specific use cases. Available through Daybreak Red access.

If you use SignalPlus for any other exchange platform, it's time to revoke all your API keys. It is generally recommend that you do not store your API keys with any 3rd party platform. 🙏 (Deribit also/first reported this.)

Images, upgraded—introducing ChatGPT Images 2.6. Now with GIFs and reference images. Plus, two new models are coming to the API with 50% faster image generation.

Hy4 from Tencent scores 35 on the Artificial Analysis Intelligence Index v5 and is the #2 model from a China-based AI lab Hy4 is the latest model from @TencentHunyuan and is the successor to Hy3 released earlier this year. It scores 35 on the Artificial Analysis Intelligence Index v5, equal with Grok 4.7 (xhigh) and 1 point behind Claude Haiku 5.5 (max). Hy4 is the #2 model from a China-based AI lab, behind only Kimi K3.1 Pro, which scores 46. Hy4 is priced at $0.834/$2.501 per 1M input/output tokens. Key takeaways: ➤ Hy4 excels at agentic coding tasks. Hy4 scores 31% on FrontierCode 2.0, compared to 32% for Claude Fable 5.1 (max). It is the leading model from a China-based AI lab, with DeepSeek V4.1 Flash 1017 (max) scoring 28% and Kimi K3.1 Pro scoring 25%. On Terminal-Bench 4.0, it scores 37%, the fourth-highest score among models from China-based AI labs ➤ Hy4 outperforms its peers on knowledge and reasoning. Hy4 achieves accuracy of 50% on AA-Omniscience v1.1, 1 point ahead of Kimi K3.1 Pro, but has a high hallucination rate of 89% compared to Kimi K3.1 Pro’s 37%. On HLE-Diamond, it scores 33% compared to Kimi K3.1 Pro’s 37% ➤ Hy4 offers a competitive Capability vs. Cost trade-off compared to to peers. With a Cost per Intelligence Index Task of $0.78, Hy4 (35) is 10-times cheaper than Grok 4.7 (xhigh, 35) and 13-times cheaper than Qwen3.8 Max (0902) (34) but almost twice as expensive as Claude Haiku 5.5 (max, 36) Other model details: ➤ Context window: 1 million tokens ➤ Pricing: $0.834/$2.501 per 1M input/output tokens. Cache reads $0.042 per 1M tokens ➤ Multimodality: Text and image input, with text output

BREAKING 🔥: GPT-6.2 Astra and GPT-6.2 Sol have been spotted on the Google Cloud Console quotas page! They are indeed coming soon! 👀

GPT Image 2.6 Sunburst and Flare by @OpenAI are here, running almost 2x faster than GPT Image 2.5. In the Text-to-Image Arena, Sunburst (Medium) is #2 with 1409 pts, and Flare (Medium) #4 with 1384 pts! By category, Sunburst landed #1 in Cartoon, Anime & Fantasy. In the Image Edit Arena, Sunburst (Medium) is #3 with 1476 pts, and Flare (Medium) #5 with 1462 pts. See leaderboard in thread. With this release, @OpenAI remains the #1 lab in both arenas and holds all top 5 Text-to-Image spots! GPT Image 2.6 lands on Arena with enhanced speed. According to @OpenAI, Sunburst (Medium) and Flare (Medium) are more than 2x faster than GPT Image 2.5 while maintaining the lead over competing labs. They report similar quality to Image 2.5, with enhanced photorealism, greater editing consistency, and a dynamic reasoning level that enables latency as low as 5 seconds. Find more details at their blog linked below. Congrats to the @OpenAI team on this release!

$1.7B in 2023 ➡️ $176.8B in 2026. Enterprise AI spend is up roughly 100x in three years, and growth is accelerating: 5.4x this year vs. 3.9x last year. @timt on why buyers aren’t pulling back. Read Menlo's 2026 State of Enterprise AI report: mnlo.vc/enterprise-ai-26

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.

Regarding criticism and suggestions about Binance’s security and access management, we accept them all. We have always believed that transparency builds trust. We read every comment one by one. Binance is not a perfect company, but a company that is constantly evolving. Criticism shows that people have expectations and standards for Binance, and these are sources of motivation for our growth. Binance has never shirked responsibility where problems exist. At present, as a trading platform, Binance needs to address three issues: 1）Will users’ funds be affected? Binance will bear full responsibility，this is what we should do. Legally, if the funding gap is large, one could file for bankruptcy and run away, but Binance operates in a “user-centric” way, with contractual spirit and a sense of moral responsibility. 2）How did the problem occur? Before publication, many professionals advised that it would be best not to include details, but we believe transparency creates trust. If this report is of reference value to other peers regarding the complexity of social engineering and the warning about AI permission boundaries, then the report is meaningful. 3）How can it be prevented in the future? We have already restructured security boundaries across departmental access management, security environments, updated devices, reset AI tools, and in accordance with zero-trust and least-privilege principles. A security audit will subsequently issue a report. Finally, we welcome more top AI security experts and technical talent to join Binance. You can DM me directly or apply on the Binance website for positions you are interested in. Joining Binance to verify this security system is the best way to validate it. 关于对币安的安全和权限管理的批评建议，我们都认。我们一直相信坦透明会建立信任，大家的评论都有一条一条看，币安不是完美的公司，但是一个一直在进化的公司，批评说明对币安有期望，对币安有标准，这些都是我们成长的动力源泉。 币安有问题的地方也从来没有推卸过责任。当下，币安作为交易平台需要解决三个问题： 1）用户的钱会不会受影响？币安会完全承担，这是我们应该做的。法律上资金缺口大可以申请破产跑路，但币安经营是“以用户中心”，是契约精神和大义。 2）问题是怎么发生的？发布前很多专业人士的意见是最好没有细节，但我们认为透明产生信任，如果这个报告在社会工程的复杂度，在AI权限边界的警示对其他同行有参考意义，那么这个报告就有意义。 3）以后如何杜绝？当前已经从各部门权限管理，安全环境，更新设备，重置AI工具，并遵循0信任、最小化原则重构安全边界，后续会由安全审计出具报告。 最后欢迎更多顶尖的AI安全专家和技术人才加入币安，您可以直接DM我本人，也可以在币安网站申请你感兴趣的岗位。加入币安来验证这套安全体系是最好的验证办法。

New: “Family Basic” subscription lineup showed up in the ChatGPT web app checkout pricing config

$2.8 billion net inflow over the last 24hrs, according to public data.

Signing in to your company's plugins one by one is annoying, so we let Claude auto-connect to them now. Team and Enterprise admins can set them up, and teammates are connected automatically the first time they use the plugins.

A hack of this scale means serious security failures occurred. No platform is unhackable, but that cannot be an excuse. Every incident exposes weaknesses that should have been prevented. Some details are embarrassing. Public explanations also simplify what appears to have been a prolonged multi month/year attack targeting hundreds of employees and systems, with a few ultimately compromised. Sharing those details, however uncomfortable, helps the wider industry learn. As a shareholder, I expect Binance to be transparent about what went wrong, take responsibility, learn from all criticism, and make meaningful improvements. 🙏

The biggest story of the year? Agents. Enterprises planned to spend 1.9x last year’s AI budget. On average, they spent 4.7x. Agents change the economics of AI. They work in parallel and keep going after people step away, so spend scales with work, not headcount. The same pattern shows up in usage. On @OpenRouter*, a platform that gives developers access to hundreds of AI models, agents now generate 80% of all tokens, about 7x human-driven usage. Agent traffic is up 37x since it overtook humans in February. The center of AI usage has moved from a person with a question to a machine with a task. More from our 2026 State of Enterprise AI report: mnlo.vc/enterprise-ai-26 *Menlo Ventures is an investor in OpenRouter.