B.AI
Sep 14, 2026

B.AI says registered users topped 2.6 million after crossing three milestones in five days

B.AI said on Sept. 14 that its total number of registered users has reached 2,600,264. According to the platform, it moved past 2.4 million, 2.5 million and 2.6 million users within just five days, marking three consecutive 100,000-user milestones over a short span. The company linked that growth to its stated ability to handle more than 1T tokens in daily throughput, which it said supports production-scale workloads. B.AI also pointed to what it described as barrier-free access to frontier models, cost-efficient computing orchestration and stable platform capacity as factors drawing more developers to the service. In its statement, the platform said usage spans early AI experimentation, production-grade deployments, single model calls and more complex Agent workflows. It added that it is continuing to connect more models, users and developers, while inviting users to access the service through chat.b.ai/chat.

1270
B.AI says registered users topped 2.6 million after crossing three milestones in five days
OpenAI
Sep 12, 2026

Users Say OpenAI’s GPT-6 Astra Feels Worse One Week After Launch

OpenAI’s GPT-6 Astra is facing a wave of complaints just a week after launch, with developers and users on X saying the model now feels less capable than it did during its debut. Several posts describe a similar pattern: faster replies, lower-quality output, and suspicion that OpenAI may have reduced how much compute the model spends reasoning before answering. Some users said side-by-side tests using the same prompts and settings produced weaker results from the current Astra than from the launch-day version, while others said they had already switched back to GPT-5.6 Sol because the higher cost was not justified. Not everyone agrees that Astra was deliberately weakened. Some users argued the model has not changed and that launch-week excitement simply faded, giving people more time to notice flaws that were always there. The debate echoes complaints that hit GPT-5.6 Sol in July, when OpenAI executive Tibo Sottiaux denied intentionally weakening the model while acknowledging experiments with reasoning effort. OpenAI has not yet issued a similar statement on Astra. The model still carries pricing of $10 per million input tokens and $50 per million output tokens, while remaining OpenAI’s first model to cross what the company describes as a critical threshold for cybersecurity risk.

1510
Users Say OpenAI’s GPT-6 Astra Feels Worse One Week After Launch
B.AI
Sep 11, 2026

B.AI to End Free Trial for GLM-5.3-Flash, Roll Out 90% Off API Pricing on Sept. 12

B.AI said its limited-time free trial for GLM-5.3-Flash is nearing its end, with a dedicated 90% discount on API calls set to take over at 10:00 SGT on Sept. 12. Under the new pricing, users will pay 10% of the original rate, with input priced as low as $0.015 per million tokens and output as low as $0.05 per million tokens. The company describes GLM-5.3-Flash as a natively multimodal large model with 320 billion total parameters and 18 billion active parameters. It also cites a hybrid sparse and linear attention architecture and a 1 million-token context window. According to B.AI, the model is designed for use cases including programming development, AI agents, long-context processing, and high-frequency API workloads. The company said it is keeping a low barrier to access and is aiming to offer developers lower-cost access to high-end compute.

1330
B.AI to End Free Trial for GLM-5.3-Flash, Roll Out 90% Off API Pricing on Sept. 12
DeepSeek
Sep 11, 2026

DeepSeek drops forced migration plan and will keep V4 Pro after Sept. 14

DeepSeek has reversed its earlier plan to force users off V4 Pro. In its latest notice, the company said the V4 Pro API will remain available after Sept. 14 and pricing will stay unchanged. It also canceled the previously announced automatic redirection of deepseek-v4-pro requests to V4.1 Flash. The move marks a clear shift from DeepSeek’s position two days earlier, when it said V4.1 Flash outperformed V4 Pro across performance, pricing, speed and total runtime, and that V4 Pro would be taken offline on Sept. 14. Under that plan, all requests sent to deepseek-v4-pro would have been routed to V4.1 Flash. That proposal drew pushback from developers. Many production environments had already tuned prompts and agent workflows around V4 Pro, raising concerns that a model swap at the backend could change output behavior. DeepSeek said it decided to keep V4 Pro "in response to broad user demand," effectively withdrawing the mandatory migration plan.

1040
DeepSeek drops forced migration plan and will keep V4 Pro after Sept. 14
B.AI
Sep 11, 2026

B.AI says registered users topped 2,500,129 after adding 100,000 in a short period

B.AI said on Sept. 11 that its cumulative registered user base had reached 2,500,129, rising from 2.4 million to more than 2.5 million after adding 100,000 users over a short period. According to the platform, the increase reflects broader adoption by developers and AI users from around the world, with more people incorporating B.AI into daily use and work routines. The company also said it is continuing to connect more models, users, and agents on the platform. In that structure, B.AI said it handles routing, metering, and settlement for each call between the model layer and the agent layer. The announcement also included a call for users to visit b.ai and join what it described as an intelligent connection network.

1460
B.AI says registered users topped 2,500,129 after adding 100,000 in a short period
Hangzhou
Sep 10, 2026

Hangzhou launches coding subsidy program with 51% off Zhipu annual plans

Hangzhou has rolled out a "citywide Coding Plan" that subsidizes Zhipu's GLM Coding Plan for eligible workers in the city, students at Hangzhou universities, and companies in Shangcheng District. For individuals, the combined subsidy reaches 44% for quarterly plans and 51% for annual plans, bringing the annual package down to roughly 49% of the original price. BlockBeats cited the Pro version as an example: the annual plan drops from 6,456 yuan to 3,163.44 yuan, while the Lite version falls from 1,416 yuan to 693.84 yuan. Each person can claim the subsidy only once and must choose either a quarterly or annual plan, with the program ending once the total subsidy pool is exhausted. Workers need to submit Hangzhou social security proof, while students must provide enrollment documentation from a Hangzhou university. Corporate support is higher but limited to companies in Hangzhou's Shangcheng District. The team annual plan receives a 55% combined subsidy, and support for a single company is capped in principle at 1 million yuan. Companies that already bought team packages can also apply for subsidies for additional seats. Official materials said the funding comes from "government subsidies, corporate concessions, and market participation."

1760
Hangzhou launches coding subsidy program with 51% off Zhipu annual plans
Solana
Sep 9, 2026

Solana to hold Solana Accelerate China roadshow across four cities in October

Solana said it will host the Solana Accelerate China roadshow from Oct. 16 to Oct. 22, with events scheduled in four Chinese cities. The company said the tour will bring together developers, founders and investors for in-person exchanges. According to the announced schedule, the roadshow will start in Shanghai on Oct. 16, continue to Hangzhou on Oct. 18, move to Shenzhen on Oct. 20, and conclude in Beijing on Oct. 22. The plan was disclosed in a post published by Solana on X.

2590
Solana to hold Solana Accelerate China roadshow across four cities in October
OKX
Sep 4, 2026

OKX launches AI Builder program with commission rebates of up to 50%

OKX has officially launched its AI Builder co-building program and opened applications, according to an official announcement. The program allows developers to connect to OKX market data, trading, and account functions through MCP, CLI, or API, then embed trading features into their own AI products. Under the arrangement, when users complete trades through a Builder project, the Builder can share in the fee revenue generated with OKX. The commission rebate can reach as high as 50%. OKX said both new and existing users’ trades are eligible to be counted toward the rebate program. The announcement outlines the access methods, the available platform capabilities, and the revenue-sharing structure tied to user trading activity completed through participating Builder projects.

1590
OKX launches AI Builder program with commission rebates of up to 50%
Total 62 items
Go topage