Applied artificial intelligence, neural systems, and smart agents.
Fusing generative AI, multi-agent frameworks, and high-throughput data processing to automate business workflows.
Kimi K3 Max
Kimi K3 records a score of 1674 on the WebDev Arena leaderboard, eighteen points below Opus 5, and at $3 in and $15 out it is priced roughly forty percent lower. The first thing to establish: K3 Max is not a standalone model in Moonshot own technical documentation.
- 1674 on WebDev Arena
- 1,048,576 token window capacity
- Cached input at $0.30
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. Change log
Kimi K3 Max is not the name of a standalone model
The entry recorded on the Arena leaderboard carries the id kimi-k3-max, but Moonshot model catalogue lists no model under that name. It lists kimi-k3. The max suffix is a configuration setting — the amount of effort the model spends on reasoning — not a separate product with a separate price.
The same pattern repeats on GLM-5.2 and on DeepSeek, and most comparison lists treat each entry as its own model. This page kept the slug that matches what users actually search for, but attached the pricing rate and the context window capacity to the base model, because the source publishing those two figures is describing the base model.
The pricing rate, and a second rate that applies only to repeated prompts
Three dollars per million input tokens and fifteen per million output. Against Opus 5 at $5 and $25, that is forty percent lower. Moonshot also specifies a second input rate: when the prompt already sits in cache, input is billed at $0.30, a tenth of the standard rate.
That $0.30 figure is not recorded in this reference pricing column, for a plain reason: no real-world workload achieves one hundred percent cache hits. If a deployment architecture carries a long fixed prefix repeated thousands of times a day, the cached rate is genuinely applicable and should be computed separately for that workload. If every request is fully fresh, three dollars is the applicable rate.
Why it holds third position rather than second in this table
Kimi sits 46 points ahead of Fable 5 on the web-dev leaderboard, yet finishes 10.7 points behind it in this reference table. Three columns build that outcome, and the largest is tooling maturity: Fable 5 takes full marks across all fifteen weight points, and Kimi takes zero, because Moonshot platform documentation lists neither an official command-line tool nor an editor extension. The other two columns are the general text leaderboard and the Iran reachability column, where Kimi carries no measurement evidence at all.
So for a deployment team that writes its own tooling layer and requires no ready-made agent SDK, those fifteen points carry no relevance, and Kimi effectively sits higher than this table indicates. The table cannot account for that variable — the deployment team can.
The column Kimi sweeps outright, and why that result is not fully reliable
Kimi context window capacity is 1,048,576 tokens — two to the twentieth power, not a rounded marketing figure of one million. That sub-five-percent difference assigns Kimi the full ten points of the context window column. Before Grok 4.5 joined this table, that same rounding margin left every other model at zero; now that Grok sets the floor of the column at 500,000 tokens, the one-million-token models draw a score of 9.1.
Min-max normalization produces exactly this behavior and is not adjusted in this report, because manually bending one column invalidates the rest of the table. But a reader should register that those ten points do not represent ten points of genuine advantage — in practice, the two window capacities are roughly the same size. A second point follows from this: the arrival of one new entry with a smaller window capacity shifted this column score for every model, without any model itself having changed.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- not checked
- Payment
- not checked
- Free tier
- no
How we checked: Moonshot publishes no supported-countries list, so this status cannot be recorded as either open or blocked. Unmeasured is a valid status within this measurement architecture, and it draws no points in the ranking.
Documented strengths
- 1674 on WebDev Arena, above Fable 5 and below Opus 5
- The lowest rate among the top three in this coding table, at $3 and $15
- The largest context window capacity in this table, 1,048,576 tokens
- A cached input rate of $0.30, for architectures with a long fixed prefix
Room for improvement and technical considerations
- Moonshot documents no model named Kimi K3 Max, so what the leaderboard measured is a configuration setting rather than a standalone product.
- It scores zero in the tooling maturity column: no official command-line tool and no editor extension in the documentation.
- Moonshot publishes no supported-countries list, so the Iran reachability column is recorded as unmeasured and draws no points.
- On the general text leaderboard it records a score of 1487, below both Opus 5 and Fable 5.
- The $0.30 rate is applicable only when the prompt genuinely hits the cache, which must be computed against the specific workload.
Technical verdict
For a deployment team that writes its own tooling layer and carries high token spend, Kimi K3 is the most sensible trade-off in this table: eighteen points behind Opus 5 at a forty percent lower rate. For a deployment that wants to pick up a model on day one along with ready-made tooling, this is not the right choice.
Frequently raised questions, with documented answers
Is Kimi K3 Max different from Kimi K3
Moonshot catalogue records only kimi-k3. Max is the reasoning-effort level, configured on that same model, so the pricing rate and the context window capacity are identical.
What is the pricing rate for Kimi K3
Three dollars per million input tokens and fifteen dollars per million output tokens. When the prompt sits in cache, input is billed at $0.30.
Is Kimi reachable from Iran
This status is unmeasured, and that is exactly what is recorded here. Moonshot publishes no supported-countries list and no direct measurement from an Iranian connection has been run. Any answer other than unmeasured at this stage would be a guess.
For coding, is Kimi or Opus 5 the better choice
On the web-dev leaderboard, Opus 5 leads at a score of 1692 against Kimi 1674. The score gap is small and the rate gap is not, so the answer turns on how much value ready-made tooling carries for the deployment.