Try Kimi at kimi.com →

Kimi AI Review
Results From Real Tasks

Is Kimi AI actually good? This review skips the marketing and gathers what happened when independent testers gave Kimi real work: coding jobs, app builds, trap tasks, desktop automation and video. Every result names its source and date.

8 independent tests Kimi K3, K3.1 Pro and Kimi Work Pros, cons and verdict
K KIMI REPORT CARDFrom published real-task tests Everyday coding Strong Websites and front-end Strong Big coding builds Good Research and long files Good Tricky edge cases Mixed Speed Slow Value for money Strong 4/5 VERDICT Great value · check its work 8 independent tests · Jun–Oct 2026
VERDICT

The short verdict

Kimi is one of the best-value AI models you can use today. On clear, everyday tasks it matches far more expensive models for a fraction of the cost. It is slower, it writes a lot, and it slips more often on tricky edge cases, so it rewards people who check its work.

4 / 5our verdict

Great value, with speed and reliability trade-offs.

≈⅓of Claude's cost

For the same three coding jobs in The New Stack's test.

4×slower

Kimi K3's total time against Claude Fable 5 in the same test.

55AA index score

K3.1 Pro ranks fourth on the Artificial Analysis Intelligence Index.

Pros

  • Matches top models on clear coding tasks
  • Strong at websites and copying designs
  • About a third of the cost of Claude per finished job in testing
  • 1M-token context for long files and codebases
  • K3.1 Pro is a big step up from K3
  • Open weights you can host yourself

Cons

  • Slow: up to four times longer than Claude
  • Very wordy, which eats into the savings
  • Weaker on hidden rules and trap tasks
  • Can miss a requirement or need follow-up prompts
  • Kimi Work is still in beta
  • Data is processed on Moonshot's servers
METHOD

How this review was made

Real tasks only

We used published hands-on tests where testers gave Kimi actual work, plus one widely used independent benchmark index. Moonshot's own launch benchmarks are not counted as results.

Every result is sourced

Each test names who ran it, when, which models were compared and links to the original. Figures are reported as the testers gave them.

Ratings are our reading

The Strong, Good, Mixed and Slow labels and the 4 / 5 verdict are our summary of these results, not a lab score of our own.

RESULTS

Results from real tasks

Eight independent tests from June to October 2026, covering Kimi K3, Kimi K3.1 Pro and Kimi Work.

01Strong

Three real coding jobs

Kimi K3 vs Claude Fable 5 · The New Stack · 20 Jul 2026

A bug fix, a refactor and a new feature in the open-source fd search tool, using the same prompts for both models.

  • Both models produced fixes that passed every test (70/70 and 264/264).
  • Kimi K3 cost $2.13 for all three jobs, against $5.98 for Fable 5.
  • Kimi K3 took about 28 minutes in total, against about 7 for Fable 5.

Result: Same quality at about a third of the cost, but four times slower.

Read the source ↗
02Mixed

GitHub issues and trap tasks

Kimi K3 vs Claude Opus 4.8 · MindStudio write-up · 25 Jul 2026

Real GitHub issues from an active codebase, each planned, built and checked, then scored out of 70. A set of trap tasks targeted known weak spots.

  • Simple issues: Kimi K3 scored within about 0.2 points of Opus 4.8's 64.3.
  • Bigger builds: Opus averaged 62.2 and Kimi K3 stayed above 60.
  • Trap tasks: Kimi K3 failed 36% of the time, against 8% for Opus.

Result: Close on clear work, much weaker on tricky, hidden-rule tasks.

Read the source ↗
03Good

Three app builds

Kimi K3.1 Pro vs Claude Opus 5.5 · AI Fire · 28 Oct 2026

The same prompts for a website copy, a task-management dashboard and a 3D racing game.

  • Website copy: Kimi K3.1 Pro won, with a more accurate replica in 12 minutes against 20.
  • Dashboard: Claude won, with clearer task details.
  • 3D racing game: Claude won, with better graphics and track design.

Result: Kimi won 1 of 3. Best at copying designs, behind on complex builds.

Read the source ↗
04Good

Visual coding: K3.1 Pro vs K3

Kimi K3.1 Pro vs Kimi K3 · Overchat · 24 Oct 2026

Two browser apps: an aquarium simulation and a low-poly 3D island with a day-night cycle.

  • K3.1 Pro wrote richer code (about 33,000 characters against 19,000) and a more detailed daytime island.
  • But it missed one requirement: the night sky had no stars, while K3 included them.
  • K3.1 Pro took 10.6 and 11.6 minutes, against 3 and 4.1 for K3.

Result: More detail and accuracy overall, but two to three times slower, and still worth checking.

Read the source ↗
05Good

A real 3D workflow

Kimi K3.1 Pro vs Claude Opus 5.5 · InfoQ developer test · Oct 2026

A developer used K3.1 Pro in a real 3D modelling workflow soon after launch.

  • K3.1 Pro produced usable 3D work.
  • It didn't finish in one prompt and needed several rounds of asking it to keep improving.
  • Claude Opus 5.5 had produced a more complete result first time in the developer's earlier test.

Result: Capable at 3D, but expect to iterate.

Read the source ↗
06Good

Kimi Work desktop tasks

Kimi Work beta · Machine Heart · 4 Jun 2026

Five jobs at launch: an industry financing list in Excel, sorting local files, posting through a logged-in browser, 30 long speeches, and a stock analysis.

  • Parallel agents built the financing list and an Excel summary.
  • The 30-speech job produced a 104-page Word file in just over 50 minutes. It hit one timeout and restarted on its own.
  • The stock analysis made 20 tool calls and wrote an 11-page report.

Result: Real end-to-end work on the desktop, with beta-stage hiccups.

Read the source ↗
07Good

Video understanding

Kimi K3 API · Kimi-AI.chat · 29–30 Jul 2026

300 fixed questions about 50 short test videos, sent through the Kimi API.

  • Kimi K3 answered 222 of 300 questions correctly.
  • That is 74.0% on this test set.
  • The test didn't cover the Kimi app, long videos or audio.

Result: Solid video understanding, with one in four answers wrong.

Read the source ↗
08Strong

Independent benchmark index

Kimi K3.1 Pro · Artificial Analysis · Oct 2026

A composite index of reasoning, knowledge, maths and coding tests, run independently.

  • K3.1 Pro scored 55, fourth overall, up 11 points from K3's 44.
  • Only Claude Opus 5.5 (58), Claude Sonnet 5.5 and Muse Wonder Full ranked higher.
  • It used 150M output tokens on the index, against an 87M median for its price tier, at $2.11 per task.

Result: Near the top on quality, but slower and wordier than most.

Read the source ↗
NUMBERS

Head-to-head numbers

What the numbers say

  • 1
    Same results, lower bill. On three real coding jobs, Kimi K3 passed every test, like Claude Fable 5, for $2.13 instead of $5.98.
  • 2
    Time is the trade-off. The same jobs took Kimi about 28 minutes, against 7 for Fable 5.
  • 3
    Traps expose the gap. On tasks built to catch it out, Kimi K3 failed 36% of the time, against 8% for Opus 4.8.
  • 4
    K3.1 Pro closes in. It scores 55 on the Artificial Analysis index, three points behind Claude Opus 5.5 and 11 ahead of K3.
Coding jobs Coding jobs Trap tasks Trap tasks Independent index Independent index TOTAL COST, 3 JOBS Kimi K3 $2.13 Claude Fable 5 $5.98 TOTAL TIME Kimi K3 28 min Claude Fable 5 7 min ✓ Same quality: every test passed on all three jobs Source: The New Stack, 20 Jul 2026 · fd Rust repo SIMPLE GITHUB ISSUES (OUT OF 70) Kimi K3 ≈64.1 Opus 4.8 64.3 TRAP TASKS FAILED Kimi K3 36% Opus 4.8 8% Near-equal on clear tasks, slips on hidden traps Source: MindStudio write-up of an independent test, 25 Jul 2026 ARTIFICIAL ANALYSIS INTELLIGENCE INDEX Claude Opus 5.5 58 Claude Sonnet 5.5 56 Muse Wonder Full 56 Kimi K3.1 Pro 55 Kimi K3 44 4th overall, +11 over K3, but very verbose Source: Artificial Analysis, Oct 2026

Cost, time, trap-task failures and independent index scores from the tests above.

SCORECARD

Kimi AI scorecard

AreaRatingWhat the tests showed
Everyday codingStrongMatched Claude on bug fixes, refactors and simple GitHub issues.
Websites and front-endStrongBeat Claude Opus 5.5 at copying a website, and faster.
Big coding buildsGoodSlightly behind Claude on complex dashboards, games and large builds.
Research and long filesGoodKimi Work produced long documents, Excel files and data reports.
Tricky edge casesMixedFailed 36% of trap tasks and can miss a stated requirement.
SpeedSlowUp to four times slower than Claude, and slower than average.
Value for moneyStrongAbout a third of Claude's cost for the same coding results.

Ratings are our reading of the results above. Kimi changes quickly, so newer versions may do better.

COST

What Kimi costs in practice

Cheaper per job, but watch the word count

Kimi's per-token prices are low, but its models think and write a lot. Artificial Analysis found K3.1 Pro produced 150 million output tokens on its index, well above the 87 million median for its price tier. That is why its cost per task, $2.11, didn't fall as far as its per-token prices did.

Even so, in the hands-on coding test Kimi K3 finished the same three jobs for about a third of Claude Fable 5's bill.

How to keep costs down

  1. Lower the thinking effort to high or low for routine tasks.
  2. Reuse long prompts so context caching cuts input costs by about 90%.
  3. Use the Batch API for jobs that can wait, at 60% of list price.
  4. Measure cost per finished task on your own work, not price per token.
Kimi API pricing →

Using the Kimi app instead? See Kimi AI pricing and is Kimi AI free?

FIT

Who should use Kimi?

A great fit for

  • Developers who want near-frontier coding at a much lower price
  • Front-end and website work, including copying a design
  • Long documents and big codebases (up to 1M tokens)
  • Teams that want open weights or self-hosting
  • Desktop automation with Kimi Work

Think twice for

  • Work where every second counts, unless you pay for Highspeed
  • High-stakes code with hidden rules and no tests to catch mistakes
  • Anyone who needs a result first time without reviewing it
  • Sensitive data your organisation hasn't approved for Moonshot's servers

Choosing a version? See all Kimi models, Kimi K3.1 Pro and Kimi vs DeepSeek.

TRY IT

Test Kimi yourself

The best review is your own. These six prompts match the kinds of tasks tested above. Run each in Kimi and in the AI you use now, then compare the results, time and cost.

Coding

Fix a real bug

Here is a failing test and the code it covers. Find the cause, fix the code rather than the test, and explain the bug in two sentences.

Front-end

Copy a design

Recreate this screenshot as a single HTML page with CSS. Match the layout, spacing and colours as closely as you can.

Research

Cited summary

Compare [3 products] on price, features and reviews in a table. Cite a source for every number, and list anything you couldn't confirm.

Trap check

Test the edge cases

Write a function that splits a bill between [n] people, rounded to cents. Then list every edge case you can think of and test each one.

Long file

Long-document check

Read this [report]. Give me the five most important facts with page numbers, then one thing that looks inconsistent.

Kimi Work

Desktop task

In Kimi Work: read the files in my [folder], build an Excel summary with one row per file and its key figures, and don't change the originals.

New to Kimi? Start with how to use Kimi AI, or try Kimi Work and Kimi Code.

LIMITS

Limits of this review

  • Tests age fast. Several results are from July 2026, before K3.1 Pro. Newer versions may do better.
  • Small samples. Several findings come from a single test with a handful of tasks.
  • Different setups. Testers used different tools, prompts and settings, so numbers aren't directly comparable across tests.
  • We're a Kimi guide. We are independent of Moonshot AI, but we cover Kimi in more depth than its rivals.

Sources

  1. Three real coding jobs · The New Stack · 20 Jul 2026
  2. GitHub issues and trap tasks · MindStudio write-up · 25 Jul 2026
  3. Three app builds · AI Fire · 28 Oct 2026
  4. Visual coding: K3.1 Pro vs K3 · Overchat · 24 Oct 2026
  5. A real 3D workflow · InfoQ developer test · Oct 2026
  6. Kimi Work desktop tasks · Machine Heart · 4 Jun 2026
  7. Video understanding · Kimi-AI.chat · 29–30 Jul 2026
  8. Independent benchmark index · Artificial Analysis · Oct 2026
FAQ

Kimi AI review FAQ

Yes, for most everyday work. In independent tests, Kimi K3 matched Claude on clear coding tasks at about a third of the cost, and Kimi K3.1 Pro ranks fourth on the Artificial Analysis Intelligence Index. It is slower than rivals and less reliable on tricky edge cases, so check its work.

Not overall. In the tests in this review, Claude was faster and more reliable on complex and trap tasks, and won two of three app builds against Kimi K3.1 Pro. Kimi matched Claude on simple coding jobs, won at copying a website, and cost much less.

Yes. Artificial Analysis scores K3.1 Pro at 55 against 44 for K3, and in Overchat's visual coding test it built more detailed apps. It can take longer per task, though, and it still missed a requirement in one test.

Often, yes. Kimi K3 took about four times as long as Claude Fable 5 on three coding jobs, and Artificial Analysis rates K3.1 Pro as slower than average. K3.1 Pro Highspeed is two to three times faster, according to Moonshot, at double the price.

Per token, yes, by a wide margin, and in The New Stack's test Kimi K3 cost $2.13 against $5.98 for Claude Fable 5 on the same jobs. Kimi models can use a lot of output tokens, though, so compare cost per finished task, not just the price list.

Yes, like every AI model. In one test Kimi K3 failed 36% of trap tasks designed to catch it out, and in another K3.1 Pro left out a feature the prompt asked for. Run tests, check facts and review its output before you rely on it.

Early tests were positive. Kimi Work built Excel lists, sorted local files, posted through a logged-in browser and wrote a 104-page document on its own. It is still in beta, and testers saw occasional timeouts.

It brings together published, independent hands-on tests and benchmarks from June to October 2026, each named and dated, rather than our own lab tests. The ratings are our reading of those results.

Kimi is made by Moonshot AI, a Beijing-based company, and your chats are processed on its servers. Don't share sensitive personal or company data unless your organisation has approved it. Kimi Work asks before changing files in its default setting.

Try Kimi on your own tasks

Start free in the Kimi app and see how it handles your real work.