Realtime AI News
StepFun's Step 5 Preview claims a global open-source top-three spot at one-eighth the cost of Claude Opus 5
According to a report on mrjjxw.com, StepFun has released Step 5 Preview, claiming it has entered the global open-source top three and that its cost per task is just one-eighth that of Claude Opus 5. The same report notes that beyond leaderboard placement, real-world deployment still requires work.
There is a new name in the open-source model race. According to a report on mrjjxw.com, StepFun has released Step 5 Preview along with two headline claims: that the model has entered the global open-source top three, and that its cost per task is one-eighth of Claude Opus 5.
Those two numbers together point at the most practical front in open-source competition today: not whether a model can rank, but what an equivalent task costs to run.
Holding per-task cost at one-eighth of Claude Opus 5 means the same budget covers many more rounds in agent workflows, batch processing and synthetic data generation.
The report keeps the caveat attached, though: beyond leaderboard results, deployment in real scenarios still requires work. That is a useful reminder that rankings depend on specific evaluation sets and configurations, while enterprises care about stability in long-horizon tasks, tool-calling accuracy and sustainable inference cost.
The Preview suffix suggests this remains a stage version, with both capability and pricing basis still subject to adjustment, so the ranking and the cost figure are best read as vendor-benchmarked interim results.
Three things are worth watching: whether independent third parties reproduce the Step 5 Preview ranking on common open-source leaderboards, what hardware, context length and concurrency assumptions sit behind the per-task cost claim, and how StepFun opens the weights in terms of licence and deployment model.
Why it matters
If the cost claim survives third-party testing, Step 5 Preview will press directly on the cost-performance baseline for open models, especially in agent and high-concurrency inference settings. The real test is whether leaderboard results convert into dependable production behaviour.
Nearby Updates
All09/21, 08:00
OpenAI details how V7 gives AI agents institutional memory
OpenAI has published a customer story on V7, describing how the company uses GPT-5.6 to turn scattered corporate files into context that AI agents can act on. The piece stresses that V7's agents are meant to complete complex work while keeping every conclusion traceable to a source.
09/21, 08:55
Zhipu rolls out its strictest data protection mechanism after data-transfer controversy
Zhipu has brought online a data protection mechanism that Lianhe Zaobao describes as the strictest in China, following a controversy over alleged covert data transfers. The move puts model providers' data handling under scrutiny, and how verifiable the mechanism turns out to be will decide whether it restores trust.
09/21, 09:07
Plugin4Shell zero-click RCE hits four AI coding agents
A report captured through Google News describes a zero-click remote code execution vulnerability called Plugin4Shell that affects four AI coding agents. Zero-click means the flaw needs no extra action from a user in the interface once its trigger condition is met.
09/21, 09:48
BYD Launches Didixia AI Super-Agent, Debuting On The Tengshi N8L
BYD has released Didixia, which it calls the automotive industry's first AI super-agent, debuting on the Tengshi N8L electric vehicle and upgrading the cockpit into an AI OS on its Xuaji 2.0 architecture. The open ecosystem platform supports A2A and MCP and already connects more than ten partner agents, with several Tengshi models set to receive OTA updates.