–
days
Cost calculator: what the 33 percent cut buys you
Three sliders. Prices are each provider’s published rate card. Cached input is the cheapest line on any rate card, so agent loops that re-send the same context save real money. Kimi K3’s rates are not published separately; they are derived from Moonshot’s 33 percent cut, so treat them as an estimate. Reasoning is always on in K3.1 Pro and bills as output, which means the output slider prices the effort level too (Metarix).
100K
60%
10K
Rates: Artificial Analysis (Pro $2/$10/$0.20, 33% below K3); Skimwire (Highspeed $4/$20). K3 rates derived at 1.5x Pro. Venice is a third-party price with no Zero Data Retention. Moonshot publishes no separate cache-hit rate for Highspeed; the calculator prices its cached share at the Pro cache rate, so Highspeed rows are a low estimate.
Cost per finished task (Artificial Analysis)
Artificial Analysis ran the model through its weighted task set and counted input, caching, reasoning, and answer tokens. The result: Kimi K3.1 Pro costs $1.98 per task, roughly equal to the older K3. It uses 55,000 output tokens per task, 14 percent more than K3 at 48,000, but its token prices are 33 percent lower, so the bill lands in the same place.
Token price is not the whole bill. K3.1 Pro spends more tokens per task than K3 does, and models use different token counts for the same job. Always measure with your own prompts before committing to a rate card.
The November 4 weights question
Moonshot has committed to publishing the full open weights by November 4, 2026. No license terms have been disclosed. If the weights land as promised, Artificial Analysis expects K3.1 Pro to become the highest-scoring open-weight model on the Intelligence Index, narrowing the gap between the best open and best closed models from 12 points to 3.
Countdown to Moonshot’s pledge
–
hours
–
minutes
–
seconds
Target: November 4, 2026 00:00 UTC. Moonshot gave no time, timezone, or download page with the pledge, so this counts to the earliest moment the promise could be kept.
One claim to keep in quarantine. Moonshot says it deliberately stripped cyberattack-related data out of mid-training and reinforcement learning for the open release, so the public weights “won’t push the frontier of offensive hacking” (Skimwire). That is the company’s own description, not an independent audit. Until the weights and a license exist, treat open-weight claims as marketing plus a deadline, not a fact.
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is Moonshot’s first model in the new K3.1 series, released October 23, 2026. It is not a new base model. Moonshot describes it as Kimi K3 pushed through far more reinforcement learning across what the company calls “hundreds of downstream domains” (izz.co.in, Metarix).
The architecture is a mixture of experts: 2.8 trillion total parameters with 104 billion active per token. It accepts text and image input and writes text. The context window is 1 million tokens (Artificial Analysis).
Reasoning is always on. You pick an effort level of low, high, or max, and max is the default. Temperature and top_p are fixed, so omit them. Output defaults to 131,072 tokens and can be raised to the full 1,048,576. One API quirk: vision input does not accept public image URLs; send base64 instead (Metarix).
Both models are live now: in the Kimi app and on kimi.ai, in Kimi Work 3.1.0 or later, in Kimi Code with the /model command, and through the Kimi API as kimi-k3.1-pro and kimi-k3.1-pro-highspeed. API keys become available after a top-up of at least $1, Zero Data Retention is available to API customers, and Kimi Enterprise admins must switch the model on; it is off by default (Unite.AI, Metarix).
How much does it cost?
| Plan | Input / 1M | Cached input / 1M | Output / 1M |
|---|---|---|---|
| Kimi K3.1 Pro | $2 | $0.20 | $10 |
| Kimi K3.1 Pro Highspeed | $4 | not published | $20 |
| Kimi K3 (derived) | $3 | $0.30 | $15 |
| Venice (third-party) | $3 | $0.30 | $15 |
Pro rates and the 33% cut vs K3: Artificial Analysis. Highspeed $4/$20: Skimwire. K3 rates derived at 1.5x Pro. Venice rates: venice.ai model FAQ; Venice runs an anonymized privacy tier, not Zero Data Retention.
The headline: K3.1 Pro is a straight price cut on top of the capability jump. Highspeed costs double for two to three times the speed (Moonshot’s claim). Venice’s $3/$15 adds a margin over Moonshot’s direct price in exchange for no vendor account.
Benchmarks: strong agents, weak on science
The table mixes two kinds of numbers. Moonshot-reported scores come from the company’s own launch materials and have not been independently verified (izz.co.in). Independent scores come from Artificial Analysis. Read the label before the number.
| Benchmark | Kimi K3.1 Pro | Rival | Reported by |
|---|---|---|---|
| AA Intelligence Index | 55 (4th overall) | Opus 5.5: 58; Sonnet 5.5: 56; Muse Wonder Full: 56; K3: 44 | Independent (AA) |
| Terminal-Bench 4.0 | 67% (independent); 65.7% (Moonshot) | Opus 5.5: 66.4%; Sonnet 5.5: 64%; GPT-6 Astra: 58.2% | Independent + vendor |
| GDPval-AA | 1,846 Elo (3rd) | Opus 5.5: 1,866; Wonder Full: 1,857; K3: 1,534 | Independent (AA) |
| AutomationBench-AA | 73% (K3: 58%) | n/a | Independent (AA) |
| Kimi Code Bench 3.0 (long-horizon) | 61.2% (K3: 55.95%) | n/a | Moonshot (vendor) |
| BenchCAD (3D via CAD code) | 92.7% with tools | Opus 5.5: 96.2%; GPT-6 Astra: 95.9% | Moonshot (vendor) |
| OfficeQA Pro v2 | 64.4 (Moonshot claims SOTA) | n/a | Moonshot (vendor) |
| Humanity’s Last Exam | 52% (K3: 47%) | n/a | Independent (AA) |
| CritPt | 25% (K3: 23%) | n/a | Independent (AA) |
| SciCode | 61% (K3: 59%) | n/a | Independent (AA) |
| Hallucination rate | 53% to 44% (AA); 23.4% to 8.3% (Moonshot) | n/a | Independent + vendor |
| Cost per task | $1.98 (about the same as K3) | n/a | Independent (AA) |
What the numbers don’t show: on science reasoning, Moonshot’s own chart puts K3.1 Pro well behind GPT-6 Astra (Terminal-Bench Science 0.1: 38.1 vs 68.1). This is a coding-and-agents model, not an all-rounder (izz.co.in).
How to try it
Get an API key from the Moonshot AI platform (minimum $1 top-up; ask for Zero Data Retention if you need it), or use the Kimi app, Kimi Work, or Kimi Code. Vercel listed both models on its AI Gateway the same day, so teams already routing agents through the gateway can point at moonshotai/kimi-k3.1-pro without a new vendor account (Metarix).
One consideration, separate from price and speed: Moonshot is a Beijing-based AI lab. Compliance teams should evaluate data residency and governance on their own terms; a benchmark launch does not cover them.
Related pages on this site: DeepSeek V4.1 Flash-1017 pricing and access guide, Mercury 3 cost and speed guide, MiniMax M3.1 Flash pricing and benchmarks, Reflection Beam benchmark explorer.
FAQ
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is Moonshot AI’s newest model, released October 23, 2026. It is a post-trained version of Kimi K3, not a new base model: a mixture of experts with 2.8 trillion total parameters and 104 billion active per token, a 1-million-token context window, text and image input, and text output.
How much does the API cost?
$2 per million input tokens, $10 per million output tokens, and $0.20 per million cache-hit tokens, which Artificial Analysis says is 33 percent below Kimi K3 on all three lines. The Highspeed variant costs $4 and $20. Artificial Analysis measured about $1.98 per finished agent task.
What is Kimi K3.1 Pro Highspeed?
A companion variant Moonshot says delivers the same capabilities at two to three times the speed, at $4 per million input and $20 per million output tokens. Moonshot publishes no separate cache-hit rate for it.
Is Kimi K3.1 Pro open weights?
Not yet. The API and app versions are live; Moonshot has committed to publishing the full open weights by November 4, 2026. No license terms have been disclosed. Once published, Artificial Analysis expects it to become the highest-scoring open-weight model on its Intelligence Index.
Is Kimi K3.1 Pro better than Claude Opus 5.5?
Not overall. On Artificial Analysis’s Intelligence Index, Opus 5.5 scores 58 against K3.1 Pro’s 55. On Terminal-Bench 4.0, Moonshot reports 65.7 percent against 66.4 for Opus 5.5, while Artificial Analysis measured 67 percent for K3.1 Pro, ahead of Claude Sonnet 5.5 at 64.
What changed from Kimi K3?
Moonshot says the gain comes from far more reinforcement-learning compute across hundreds of downstream domains, not a new base model. The Intelligence Index score jumped from 44 to 55; GDPval-AA Elo rose 312 points to 1,846; AutomationBench-AA went from 58 to 73 percent; and token prices fell 33 percent across the board.
How do I try Kimi K3.1 Pro?
In the Kimi app and on kimi.ai, in Kimi Work 3.1.0 or later, in Kimi Code with the /model command, and through the Kimi API as kimi-k3.1-pro or kimi-k3.1-pro-highspeed. It is also on third-party APIs such as Venice and the Vercel AI Gateway.
Does Kimi K3.1 Pro hallucinate less?
On Artificial Analysis’s AA-Omniscience Index the score rose from 20 to 30: the hallucination rate fell from 53 to 44 percent while accuracy rose from 48 to 51 percent. Moonshot’s own internal stress test reports claim-level errors dropping from 23.4 percent for K3 to 8.3 percent; that is a company claim, not independently verified.
Why does November 4 matter?
Moonshot committed to publishing the full open weights by November 4, 2026. Independent data from Artificial Analysis suggests the release would become the highest-scoring open-weight model on the Intelligence Index, narrowing the open-versus-closed gap from 12 points to 3.
Sources
- Artificial Analysis: Kimi K3.1 Pro launch benchmarks (Intelligence Index 55, $2/$10/$0.20 pricing, $1.98/task, GDPval-AA 1,846, Terminal-Bench 67%)
- izz.co.in (vendor benchmark table with verification caveats, 2.8T/104B MoE, science-reasoning gap)
- Unite.AI (release details, “most capable 3T-class” wording, Zero Data Retention, Enterprise opt-in)
- Metarix (API model IDs, effort levels low/high/max, fixed temperature, vision base64 quirk, Vercel AI Gateway)
- Venice model FAQ ($3/$15 third-party pricing, anonymized privacy tier, not open-source today)
- Skimwire (Highspeed $4/$20, weights by Nov 4, cyberattack-data-stripping claim)
- Oossa (AA 55, 4th place, pricing summary)
Ask an AI about this guide:
Agent-readable summary: /kimi-k3-1-pro/llms.txt