¥3.9 for Unlimited Tokens! The iFlytek Xingchen Coding Plan Just Got a Full Upgrade
Shrimp Farming Is the New Developer Daily
“Raising shrimp” is quickly becoming a developer’s new daily routine.
But once you really get into it, most people hit two very real problems:
First, the “shrimp feed is expensive” — every API call, every conversation turn, every debugging session is a cost.
Second, the “shrimp gets stuck” — the moment your request volume spikes, you get rate-limited, and your agent just throws in the towel.
Developers end up “using AI frugally” instead of “fully unleashing AI.”
Today, the iFlytek Xingchen Coding Plan is getting a major upgrade.
Beyond a complete overhaul of the billing model, it’s the first to bring in Zhipu’s newest flagship model, GLM-5.1, along with other top-tier models like Qwen3.5-397B-A17B — letting developers “raise shrimp” more efficiently and freely, and build more AI applications.
¥3.9 Unlimited: From Token Anxiety to Worry-Free Usage
When “raising shrimp” or doing AI-assisted coding, the thing developers dread most is rate limiting and surprise bills from sudden high-concurrency workloads.
With this upgrade, billing has shifted from per-token to per-request.
On the surface it’s just a change in billing unit, but for developers it’s a genuine paradigm shift — you stop obsessing over “how many resources did I burn” and start thinking about “how many requests did I complete.”
▲ The upgraded Coding Plan tiers
At the same time, targeting students, independent developers, and newcomers to AI coding, the original “Starter” tier has been upgraded into “Worry-Free,” which costs just ¥3.9 and comes with unlimited requests.
No matter how much work you pile onto your “lobster” in a single day, or how many requests you fire off, you’ll never be brutally rate-limited for blowing past a request cap again. That’s true “shrimp-farming freedom.”
For the “Pro” and “Efficient” tiers used by professional developers and heavy enterprise AI users, the new plans adopt a multi-dimensional flow-control strategy built on a 5-hour sliding window, weekly resets, and monthly resets — not just a daily token tally.
In simple terms, it balances short sprints and long-term stability: your “lobster” gets explosive burst throughput and concurrency during a coding sprint, while overnight automated jobs can’t let costs spiral out of control. Compute allocation becomes smoother and smarter.
GLM-5.1 Joins: Free Scheduling Across Flagship Models
If the number of requests determines how many times you can “command” your shrimp, then the model determines just how capable the shrimp you raise really is.
Alongside the billing overhaul, the Coding Plan has also expanded its model lineup to cover a wider range of scenarios.
On top of the existing models, the plans now support:
GLM-5.1:
Zhipu’s new-generation flagship base model, with significant gains on long-horizon tasks. It can work autonomously for up to 8 hours and deliver engineering-grade results end-to-end, with overall performance on par with Claude Opus 4.6.
If you have high-intensity needs like large-scale project refactoring or building complex agents, GLM-5.1 is the most worthwhile new addition to the Coding Plan to pick up first.
Qwen3-Coder-Next:
An efficient mixture-of-experts (MoE) model purpose-built for coding agents — 80B total parameters with only 3B activated — focused on long-horizon, multi-tool, interactive real-world coding tasks.
The go-to choice for everyday code completion and lightweight debugging.
Qwen3.5-397B-A17B:
A natively multimodal vision-language model covering 201 languages, with outstanding performance in code generation, agent reasoning, and multimodal understanding.
Especially well-suited to mixed image-and-text inputs or multilingual scenarios.
With these new models, developers get far more “weapons” to choose from inside a single plan.
Reach for the strongest model on complex algorithms, switch to the most efficient one for daily completion — true multi-model free dispatch.
One More Thing: Skills Give Your Lobster Superpowers
Compute and models are just the “feed” for raising shrimp — powerful Skills are the “claws” that give your lobster its soul and superpowers.
Recently, the iFlytek Open Platform has also packaged three excellent AI capabilities into Skills: ultra-realistic voice synthesis, OCR LLM, and offline audio transcription.
With the OCR LLM, you can have your shrimp automatically read an invoice, recognize tables, and reorganize them into structured data — saving you the tedious manual entry. With offline audio transcription, it can turn a meeting recording into minutes and even auto-extract the key points.
With ultra-realistic synthesis, it can literally “speak,” generating voice replies with emotion.
Once these three Skills are connected, your lobster is no longer a text-only shrimp trapped inside a terminal — it can truly “hear, speak, and see,” and interact with the real world.
The Bottom Line
For every “shrimp farmer,” this Coding Plan upgrade means your “lobster” doesn’t just eat enough — it eats well. And a well-fed lobster works all the harder.
When cost is no longer a barrier and compute is within reach, every developer can incubate AI-native applications with even greater commercial value.
How to Subscribe
On PC, head to the Xingchen MaaS platform’s official website (https://maas.xfyun.cn/packageSubscription?inviteCode=MAAS-C4669BE4) to subscribe to the new Coding Plan.
Alternatively, register first, then click “Invite Notification” in the top-right corner and enter the invite code:
Invite code: MAAS-C4669BE4