Kimi K3.1 Pro · Media / benchmark · Personal experience
In the same Kimi K3.1 Pro discussion thread, johnmlussier said that the $99/month plan hit its five-hour quota limit within 47 minutes in a single session; S-E-P, by contrast, said that running K3.1 Pro and K2.8 Preview for eight hours used only 22% of the weekly quota. Both are personal reports and cannot support a unified conclusion about quotas, billing, or model capabilities.
In the same Kimi K3.1 Pro discussion thread, johnmlussier said that the $99/month plan hit its five-hour quota limit within 47 minutes in a single session; S-E-P, by contrast, said that running K3.1 Pro and K2.8 Preview for eight hours used only 22% of the weekly quota. Both are personal reports and cannot support a unified conclusion about quotas, billing, or model capabilities.
Tasks this can help assess: Early user feedback on Kimi subscription quotas, long-running code generation, and sub-agent workflows.
Tasks this should not be generalized to: The quotas of all accounts, overall subscription value, actual billing, model capabilities, or the stability of different tasks.
Model versions covered: The parent post discusses Kimi K3.1 Pro and K3.1 Pro Highspeed; the relevant reports explicitly mention K3.1 Pro, while S-E-P also used K2.8 Preview.
Test environment or client: Subscription plans; the client, Agent harness, account status, and server-side configuration were not specified. S-E-P only said they used codegen (subagents) on an RE project.
Reasoning levels and parameters: Not specified. S-E-P asked whether K3 should be set to the highest thinking level, but did not describe johnmlussier's setting or receive an answer.
These are personal usage reports from the same Hacker News parent thread, not a controlled evaluation. The parent post is titled “Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed” and links to an official Kimi announcement. johnmlussier reported on their $99/month plan and a single session; S-E-P responded with their own K3.1 Pro + K2.8 Preview code-generation task; SwellJoe also reported their experience with a $19/month plan in the thread.
There was no publicly available unified prompt, task list, effort setting, token count, client, account quota panel, billing detail, retry record, or independent acceptance criterion.
| Reporter | Verifiable settings | Result | Evidence boundary |
|---|---|---|---|
| johnmlussier | $99/month; one session; explicitly said they did not use swarms | Hit the five-hour quota limit within 47 minutes | Model selection, thinking level, tokens, client, and account status were not specified |
| S-E-P | K3.1 Pro + K2.8 Preview; codegen (subagents); RE project | Ran for eight hours; the original text says it “only used 22% of my weekly” | 22% is the used proportion; the total weekly quota, task breakdown, and call logs were not provided |
| SwellJoe (another user) | $19/month; experiments and benchmark runs | A single task exhausted the entire five-hour limit and consumed a large portion of the weekly quota | The comment did not clearly state the model version, so it cannot be treated as a K3.1 Pro test |
Parent post: Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed, by Jowsey, HN item 50212065.
Parent post URL: https://news.ycombinator.com/item?id=50212065; the target comment is associated with this parent through parent.
johnmlussier's comment is item 50212563; the original text explicitly says $99 a month plan, within 47 minutes with one session, and limited by 5-hour quota.
S-E-P's reply is item 50217757; the original text explicitly says they used k3.1 Pro with K2.8 preview for codegen (subagents), ran it for eight hours, and “only used 22% of my weekly”.
S-E-P then asked whether the highest thinking level was being used, or whether all tasks used only K3; this was a question, not a confirmed runtime configuration.
SwellJoe's reply is item 50214102, reporting a $19 monthly plan, a single task exhausting the five-hour limit, and a large portion of the weekly quota; this was another user's experience.
These replies show two opposing quota experiences: one $99/month user said a single session quickly triggered the five-hour limit, while another user said that a long task mixing K3.1 Pro and K2.8 Preview used only 22% of the weekly quota. The difference may be related to the model combination, task, client, thinking level, sub-agents, account, or time window, but the original posts do not provide enough information to attribute it.
“22% per week,” in S-E-P's original wording, means 22% used, not 22% remaining. This does not reveal the total weekly quota or allow the remaining time or cost to be calculated. The thread provides no prompt, effort setting, token count, client breakdown, billing information, or model-capability measurement, so this article does not infer overall quotas, actual billing, performance rankings, or K3.1 Pro's general long-task capability.
To verify these claims, fix the same subscription tier, model combination, client or harness, prompt, thinking/effort setting, number of sub-agents, task repository, runtime window, and stopping rule, and record the five-hour window, weekly used proportion, input and output tokens, tool calls, and billing.
K3.1 Pro alone, K3.1 Pro + K2.8 Preview together, and subagents enabled should be tested separately; each configuration should retain a complete usage panel and run log. Without these conditions, at most one can reproduce an observation from a particular account; the two community reports cannot be combined into a unified quota conclusion.
The figures, task set, reasoning tier, and client conditions apply only to the listed source and collection snapshot. Different versions, harnesses, or providers must not be compared directly; undisclosed parameters remain unknown.
For a reproduction, fix the model version, provider or client, reasoning tier, tools, task-set version, sample count, and collection date, and record failures, retries, and human corrections. Full steps are in the source notes below.
Hacker News · johnmlussier; S-E-P and SwellJoe in the same thread · Original publication date 2026-10-23 · Site edit date 2026-10-24
Open original sourceKimi K3.1 Pro
Download the Tabbit client to check model access
