CCH HQ

One OPEN SOURCE gateway to align your AI ecosystem.

[CCH HQ @hq_cch]

Hi, here is CCHP,
One gateway to align your AI ecosystem.

CCHP 致力于成为体验优先的全功能 LLM 网关。

我们正努力在今年秋天在 CCH-HQ 组织下以 AGPL 协议正式开源 CCH Plus 项目。

[CCH HQ @hq_cch]

DeepSeek (@deepseek_ai) opened an internal beta of an intermediate DeepSeek V4.1 Flash build. DeepSeek says the version uses a new model architecture, is natively multimodal, and is stronger, faster, and lower-cost.

Keep the existing base_url and set the model name to deepseek-v4.1-flash-expires-on-0910. Billing temporarily matches deepseek-v4-flash, with a limit of 20 concurrent requests per account.

[CCH HQ @hq_cch]

NVIDIA (@nvidia) agreed to acquire Hugging Face for $12,930,300,000.

The first six digits, 129303, are Unicode code point U+1F917 — the 🤗 emoji. Written as the hex color #129303, they also make a green very close to NVIDIA’s signature color.

https://t.co/Y00BQsmLJh

[CCH HQ @hq_cch]

OpenAI released GPT-6 Astra on Sept. 3, 2026, calling it a generational leap and saying it may mark the arrival of AGI. In a press briefing, co-founder Greg Brockman closed with “Welcome to the AGI era.”

OpenAI says Astra was its largest-scale training run yet — more than 100,000 GPUs at Stargate in Texas — and that other models played a major role supervising training for the first time.

Access starts with limited Daybreak Access institutions, then rolls out over the coming days to ChatGPT Plus, Pro, Business and Enterprise, the API (gpt-6-astra), AWS Bedrock and Microsoft Azure. API Standard is $10 / $50 per million input/output tokens; Fast is up to about 2.5× Standard speed at 2× the price. OpenAI had already named Astra the first Critical cybersecurity model under its Preparedness Framework.

[CCH HQ @hq_cch]

Reportedly, OpenAI may move up its Astra model launch to Sept. 3, 2026.

A day earlier, OpenAI said Astra is the first model to meet the Critical cybersecurity capability threshold under its Preparedness Framework — meaning, with the right tools and access, it can find previously unknown flaws and develop exploits across many well-protected systems without step-by-step human guidance. The company still frames public availability as "soon," with advanced cyber access more limited.

https://t.co/hMfICNSzaV

[CCH HQ @hq_cch]

Google (@GoogleDeepMind) released Gemini 3.8 Flash on Sept. 2, 2026. Built on Gemini 3.7 Flash, it targets software engineering and agentic knowledge work, with adjustable effort. It accepts text, image, audio and video, with a 1 million-token context and up to 64K output tokens.

On Google's official comparison table, DeepSWE v1.1 rises from 65.3% on 3.7 Flash to 71.0%, and Terminal-Bench 2.1 from 85.8% to 89.4%. Vals Finance Agent v2 (61.4%), Harvey's Legal Agent all-pass rate (10.0%), and HLE-Verified (54.9%) lead the table. Versus Claude Opus 5 it can match or beat several finance, legal, long-video, and Terminal-Bench 2.1 scores, but still trails on Terminal-Bench 4.0 (19.1% vs. 51.8%) and OSWorld-2.0 (59.0% vs. 75.4%).

API pricing matches 3.7 Flash's introductory rates: $0.75 per million input tokens and $3.75 per million output tokens through Dec. 31, 2026, then $1.50 / $7.50 from Jan. 1, 2027.

https://t.co/cfQJv400FO

[CCH HQ @hq_cch]

Anthropic (@AnthropicAI) released Claude Fable 5.1 and Claude Mythos 5.1 on Sept. 2, 2026. They are the same underlying model with different safeguards: Fable 5.1 is generally available; Mythos 5.1 is limited to trusted access for cybersecurity and life sciences work.

Both have a 1 million-token context and up to 128K output tokens. API pricing stays at $10 per million input tokens and $50 per million output tokens, while cache reads drop to $0.25 per million tokens — 75% below Fable 5. Anthropic says that cuts typical workloads by about 25%, and highly agentic tasks by up to about 45%.

https://t.co/kjvnRcV4iy

[CCH HQ @hq_cch]

Tencent Hunyuan (@TencentHunyuan) open-sourced Hy4 preview on Aug. 28, 2026: a mixture-of-experts (MoE) model with 770B total parameters, 49B activated per token, and a 1 million-token context. Tencent says it sits at the open-source frontier for coding, office, and science work.

https://t.co/SgiC7di0QU
https://t.co/jzHZmpAQlx

[CCH HQ @hq_cch]

https://t.co/rB7ug5x9xp (@Zai_org) released GLM-5.3-Flash on Aug. 26, 2026, the first natively multimodal model in the GLM-5 series, with 320B total parameters and 18B activated per token. On Artificial Analysis Intelligence Index v4.1.1 it scores 57 at $0.045 per task (discounted).

https://t.co/rB7ug5x9xp says it outperforms GLM-5.2 and approaches Claude Opus 4.8 on coding and agentic benchmarks. On in-house https://t.co/rB7ug5x9xp Code Bench v1.0 at max effort it scored 29.0 versus 29.5 for Opus 4.8. Before launch it ran anonymously as ox-alpha on OpenCode and OpenRouter, becoming the week's most popular model, with that traffic served on Chinese AI chips.

https://t.co/Y5DaUu87Ov
https://t.co/oMb9pDr4fA

[CCH HQ @hq_cch]

Alibaba's Qwen team (@Alibaba_Qwen) released Qwen3.8-Flash-Next on Aug. 26, 2026: an open-weight multimodal mixture-of-experts (MoE) model with 125B total parameters and 6B activated per token. Qwen calls it an experimental preview of the Qwen4 architecture.

[CCH HQ @hq_cch]

Every model in the world, at a glance.

CCHP 独家维护的云端价格表现已全新升级，带给你更丰富准确的模型数据，更快的更新频率，以及一个更实用的模型广场。

前往 CCHP 官网，开启全新体验⬇️
https://t.co/Tsp10RAJVq

*new-api 兼容的价格表格式即将支持。

[CCH HQ @hq_cch]

CCH Plus 现已开放预约。

即日起，您可于官网 https://t.co/SHbYkHbYmv 加入 Waitlist，后续即可第一时间收到公测通知。

本月内， CCH Plus 还将于 Telegram 群聊不定时开放内测名额，敬请期待。

[CCH HQ @hq_cch]

Anthropic 被指在 Claude Code 2.1.91 版本后内置了一份与中国关联的 Domain List，作为开源的、不提供任何逆向与反代功能的 LLM Gateway 项目，claude-hub-hub[.]app 赫然在列，令人震惊（ https://t.co/AQFcaSItXH

[CCH HQ @hq_cch]

Just wanted to say a huge thanks to @OpenAI for supporting CCH with 6 months of ChatGPT Pro!

I'll be trying to use Codex more frequently to build CCHP now—hopefully, this speeds up the release of CCHP! 🚀

[CCH HQ @hq_cch]

@jwangkun 感谢支持，我们会努力在达到我们心目中的标准后尽快开源～

[CCH HQ @hq_cch]

Hi, here is CCHP,
One gateway to align your AI ecosystem.

CCHP 致力于成为体验优先的全功能 LLM 网关。

我们正努力在今年秋天在 CCH-HQ 组织下以 AGPL 协议正式开源 CCH Plus 项目。

[CCH HQ @hq_cch]

@Jami1024 收到，持续打磨 ing！

[CCH HQ @hq_cch]

@myluke1 hi，CCHP 目前还在封闭内测阶段，项目仓库预计在今年秋天公开🥰

[CCH HQ @hq_cch]

6/6 CCH Plus 不是一款用来“收割”社区的产品。

在 AGPL 许可证允许的范围内，任何人都将可以在正式发布后免费部署、使用、分发 CCH Plus，或使用它进行盈利性活动。我们始终坚信，开放与包容是开源社区的根基，只有借助开源社区的力量， CCH Plus 才能够获得持续成长，成为社区可信赖的 LLM 网关。

[CCH HQ @hq_cch]

5/6 更灵活的插件接入能力。

在未来，CCH Plus 将开放一系列插件能力，插件代码所有权由社区贡献者所有，可选不遵循 CCH Plus 主项目的分发许可证，为社区贡献提供更灵活的选择。

[CCH HQ @hq_cch]

4/6 让掺假无处遁形。

CCH Plus 通过智能的缓存命中率回测算法和模型探测算法，为你找出 Token 使用量异常的供应商及牛头不对马嘴的返回。

我们深知部分探测方式对账号风控的影响，因此 CCH Plus 将不会使用任何有害提示词对模型进行探测，也不会在未经用户主动授权的情况下主动发起生成请求。

[CCH HQ @hq_cch]

4/6 完整的商业功能。

我们希望任何人都可以轻松地搭建一个属于自己的OpenRouter。CCH Plus 将继承 CCH 的多维度限额能力，并提供包括充值、套餐、邀请、分销、兑换在内的完整商业功能。