Alibaba just made its steepest coding-model discount run exactly when American developers are at their desks, and it's not the only Chinese lab racing prices toward zero.
Qoder, the agentic coding platform built on Alibaba's Qwen models, quietly rewrote what off-peak pricing means. Starting this week, the credit multiplier for Qwen3.7-Max drops from 0.5x to 0.1x, an 80 percent price cut, and Qwen3.7-Plus falls from 0.1x to 0.04x, a 60 percent cut. The discount window runs from 14:00 to 00:00 UTC, which lines up almost exactly with 10am to 8pm US Eastern time.
The move builds on a broader price war among Chinese AI labs that intensified after Lunar New Year, when Alibaba began undercutting Western coding subscriptions across the board.
Alibaba did not bury the intent in fine print. Qwen's own account posted on X that "if you're in the Americas, here's the twist: off-peak covers most of your workday," according to reporting from the South China Morning Post. This isn't a discount for idle servers overnight in Beijing. It's built around the hours American engineers are actually typing.
The cuts stack on prices that were already low. Qwen3.7-Max lists at $2.50 per million input tokens and $7.50 per million output tokens, already discounted to $1.25 and $3.75 under a standing promotion, according to VentureBeat. Layer the new off-peak multiplier on top and the effective daytime rate drops further still. Qwen3.7-Plus, the smaller multimodal model that handles text, video and image inputs, runs $0.40 and $1.60 per million tokens even before any discount.
Alibaba Cloud also sells a flat monthly Coding Plan, roughly $10 for the Lite tier and $50 for Pro, bundling up to 90,000 requests a month. Since February the plan has expanded beyond its own models to include GLM-5, MiniMax M2.5 and Kimi K2.5, all switchable inside Claude Code, Qwen Code, Cline and OpenClaw rather than requiring developers to abandon the tools they already use. Cursor Pro, GitHub Copilot Pro and Windsurf Pro all sit between $10 and $20 a month for comparison. Alibaba isn't competing on features here. It's making the meter nearly disappear.
Zhipu AI's Z.ai is running a version of the same play. Its GLM-5.2 model prices at $1.40 per million input tokens and $4.40 per million output tokens, against Claude Opus 4.8's $5 and $25 and GPT-5.5's $5 and $30, a gap of roughly five to seven times on output cost alone. The capability gap has mostly closed too. GLM-5.2 scores 62.1 on SWE-bench Pro against GPT-5.5's 58.6, and it trails Claude Opus 4.8 by less than two points on FrontierSWE and MCP-Atlas. Zhipu released the weights under an MIT license, so any team willing to run its own infrastructure can skip the API bill entirely.
The pressure is already showing up inside companies that can afford anything. Microsoft told engineers across its Windows, Microsoft 365 and Surface divisions to stop using Claude Code by June 30, according to a report from Windows Central, after token costs for power users reached between $500 and $2,000 per engineer per month. Those teams were pushed back to GitHub Copilot CLI. Anthropic's own pricing tells a similar story: Claude Code runs at $2 and $10 per million tokens through August, rising to $3 and $15 after that.
Frankly, that's the number that should worry Cursor, Copilot and Anthropic more than any benchmark score.
Alibaba has run this exact playbook before. It undercut AWS and Google Cloud on compute pricing for years before its cloud division turned a real profit, betting that developers who build habits on cheap infrastructure rarely migrate once the discounts taper off. Coding assistants are a smaller business than cloud infrastructure, but the mechanism is identical. Subsidize the tool a developer touches every hour of the workday, then monetize the enterprise contracts and custom model deals that follow once a team has standardized on Qwen or GLM.
None of this settles whether Chinese open models can match the frontier on the hardest engineering problems. But for the refactors, tests and boilerplate that fill most of a developer's week, the price gap is now wide enough that ignoring it costs a team real money every single month.
Also read: Kling AI nears a $3 billion round as Tencent buys into a rival's video bet • Google's Electricity Demand Jumped 37 Percent in 2025 on AI Buildout • A Former Goldman Analyst's New AI Startup Just Raised $22 Million
