Kimi K3: Moonshot's 2.8T Flagship, 1M Context, Launched July 16

TL;DR

Published launch day, July 16, 2026. This page separates what Moonshot has confirmed from what is still leak noise, and gets updated as the model card, license, and independent benchmarks land.

Kimi K3 is Moonshot AI's flagship mixture-of-experts model, launched July 16, 2026. It has 2.8 trillion total parameters, a 1M-token context window, native visual understanding, and a new architecture: Kimi Delta Attention (hybrid linear attention) plus Attention Residuals. Moonshot reports 81.2 on FrontierSWE and 88.3 on Terminal-Bench 2.0, and positions K3 second only to Claude Fable 5 and GPT-5.6 Sol on GDPval-AA v2 (1687 Elo vs Opus 4.8's 1600). Weights are not out yet.

What Is Kimi K3?

Kimi K3 is the successor to Moonshot AI's K2 line (K2.6 shipped April 2026 at 1T total / 32B active parameters; K2.7 Code followed in June and reached general availability inside GitHub Copilot). K3 nearly triples the total parameter count to 2.8T, extends context from 200K to 1M tokens, and adds native visual understanding for images and video. Thinking is always enabled, with reasoning traces exposed as separate deltas in the streaming API.

The launch is API-first. The model is kimi-k3 on Moonshot's OpenAI-compatible platform, and the consumer Kimi app exposes two tiers, K3 Max and K3 Cluster Max, for logged-in users. A recharge promotion (10-30% bonus credits on API top-ups) runs from July 15 through August 11, 2026. Moonshot is reportedly raising at a $31.5 billion valuation on the back of the release, up from $20 billion in May 2026.

Before the official launch, a beta checkpoint codenamed "Kivine" ran anonymized on LM Arena, where early testers flagged two traits that carried into release coverage: elaborate, visually rich generations (interactive 3D scenes were a repeated example) and long runtimes on hard agent tasks.

Confirmed vs Leak Noise

K3 leaked for two days before it shipped, so the coverage mixes first-party facts with estimates. Here is the split as of launch day:

Architecture: Kimi Delta Attention at 1M Context

K3 is the first Moonshot flagship built on Kimi Delta Attention (KDA), the hybrid linear attention mechanism from the Kimi Linear paper (arXiv 2510.26692, October 2025). The design interleaves KDA linear-attention layers with periodic full-attention layers in a 3:1 ratio: three linear layers handle local sequence structure cheaply, one full-attention layer preserves global information flow. At matched scale in the paper, this cut KV-cache memory by up to 75% and delivered up to 6x decoding throughput at 1M-token context while matching or beating full-attention baselines on short-context, long-context, and RL-style post-training tasks. K3 pairs KDA with what Moonshot calls Attention Residuals.

The architecture is why the pricing is flat. Serving a 1M-token context with full attention means the KV cache, not the weights, dominates memory at long sequence length, which is why several competitors charge a long-context premium. Cut the cache 75% and the premium disappears from the cost structure. Whether serving economics at 2.8T actually work at $15/M output is Moonshot's bet.

What Moonshot has not published: active parameters per token, expert count and routing, vocabulary, or training data scale. K2.6's card documented 1T total / 32B active; leak coverage guesses K3 at 40-60B active, roughly double. Until a model card lands, the 2.8T total is the only first-party parameter figure.

Benchmarks: Strong Numbers, All Self-Reported

Every K3 score below is Moonshot-reported. Launch-day skepticism on Hacker News centered on exactly this: open-weight-lineage models keep posting stunning vendor benchmarks, and the community suspects benchmark leakage into training data. Treat the table as the vendor's claim, not settled fact; independent harness results take a few weeks to land.

The BrowseComp claim deserves its own note: 91.2 in a single-agent setup with no context compression or context-management tricks. If that replicates, it means the 1M window plus KDA is doing the work that multi-agent orchestration and compaction pipelines normally do on long-horizon research tasks.

Early independent signal

The first day of independent testing points one tier lower than the vendor tables. Early Arena Elo aggregation puts K3 around 1486 on text and 1530 on coding, and a day-one independent evaluation concluded K3 sits "one tier away" from Fable 5 and GPT-5.6 Sol rather than at parity. Developers did consistently praise one thing the closed frontier does not offer: fully exposed reasoning traces, which one Hacker News commenter called "far, far more informative" than Fable's opaque summaries for debugging agent behavior.

The Token Economics: Always-On Thinking at $15/M

K3's pricing reads mid-tier until you account for how it generates. Thinking mode cannot be disabled, and reasoning_effort supports only max at launch (Moonshot says more levels are coming). So every request pays for a full reasoning trace at $15/M output tokens. This is the same failure mode that makes cheap-per-token models expensive per task: GLM-5.2 burns roughly 43K output tokens per Artificial Analysis Index task at max effort, and K3 launches with max as the only option.

Day-one Hacker News math framed it directly: $3/$15 with $0.30 cache hits matches Anthropic's Sonnet-series pricing, which is expensive for a model from the open-weight lineage, and commenters noted GLM-5.2 delivers similar coding quality at roughly one-third the per-token price. The counterpoint from K3's defenders: if the GDPval-AA positioning holds under independent testing, Sonnet-tier pricing for near-Fable capability is underpriced, not overpriced. Both can be true; which one matters depends on whether your workload needs the top tier.

The flat 1M-context pricing is a genuine differentiator. Anthropic charges a premium above 200K input tokens and Google tiers Gemini pricing by context length; Moonshot charges $3/M whether you send 4K or 900K tokens. For long-context-heavy workloads (repo-scale analysis, long agent traces, document piles) that flatness changes the calculus more than the headline rate does.

Kimi K3 API: Pricing and How to Call It

The API is OpenAI-compatible: base URL https://api.moonshot.ai/v1, model kimi-k3. It supports streaming with separate reasoning and content deltas, structured JSON output with strict schema enforcement, tool calling with dynamic loading, vision inputs (base64 or uploaded file IDs), and a partial mode for prefix continuation. max_completion_tokens defaults to 131,072 and goes to 1,048,576.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.moonshot.ai/v1",
    api_key="YOUR_MOONSHOT_API_KEY",
)

resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[
        {"role": "user", "content": "Trace this race condition and propose a fix."},
    ],
)
print(resp.choices[0].message.content)

Sources: Moonshot platform pricing (flat pay-as-you-go, no context-length tiering); OpenRouter model listing, which notes effective blended cost often lands 60-80% below list because of prompt caching. One upstream provider serves the OpenRouter route at launch. Prices move fast in launch weeks; check the provider page before committing volume.

Where Are the Weights?

Not published. As of July 16 there is no Hugging Face model card, no license file, and no download for K3. That matters because the K2 family set a strong precedent: K2, K2.5, K2.6, and K2-Thinking all shipped open-weight under a Modified MIT license, and K2.6's card documented its 1T-total / 32B-active configuration. The community expectation that K3 follows is reasonable but unconfirmed, and leak coverage pointing at a Q4 2026 weights release is speculation.

If the weights do land, self-hosting will be a different weight class than anything in the K2 line: at 2.8T total, a BF16 footprint is roughly 5.6TB before KV cache, which puts even multi-node H200 clusters into careful-planning territory and makes quantized or distilled variants the realistic path for most teams. For comparison, GLM-5.2 at 753B is already a 1.5TB BF16 deployment that needs 8x H200-class GPUs for a single node.

Kimi K3 vs GLM-5.2 and DeepSeek V4

The relevant comparison set is the open-lineage frontier: GLM-5.2 (weights out, MIT), DeepSeek V4 (weights out), and K3 (weights promised by precedent, not yet delivered). The honest summary on day one: K3 posts the strongest vendor-reported agentic numbers of the three, at 2-10x the per-token price, with the least independent verification and no self-hosting option yet.

Benchmark caveat: the Terminal-Bench versions differ (Moonshot reports 2.0, Z.ai reports 2.1), and every number in the table is self-reported by its vendor, so cross-model deltas of a few points are noise. What is not noise: the price gap. DeepSeek V4 Flash on Morph costs $0.139/M input and $0.278/M output, roughly 2% of K3's output rate, and GLM-5.2 runs at about a third of K3's per-token price with weights you can hold. K3's case rests on the top-tier capability claims surviving independent testing.

For the previous Moonshot generation, see Kimi K2.5 and agent swarms. For the models Morph serves on custom codegen kernels, see Open Source Models: GLM-5.2, MiniMax M3, Qwen 3.5 397B, and DeepSeek V4 Flash.

Kimi K3: Pros and Cons

When to Use Kimi K3

Use K3 when the workload is long-horizon and context-heavy: single-agent research over large document sets (the BrowseComp setup), repo-scale analysis that genuinely needs several hundred thousand tokens in one window, or multimodal tasks where GLM-5.2 and DeepSeek V4 are disqualified for lacking vision. The flat 1M pricing makes it the cheapest way to actually use a million tokens of context in one call, even though its per-token rate is the highest of the open lineage.

Skip it, at least until independent benchmarks land, when the workload is high-volume codegen where per-task cost dominates. At $15/M output with always-on max reasoning, an agent loop that runs thousands of tasks a day costs an order of magnitude more on K3 than on DeepSeek V4 Flash or MiniMax M3, and the vendor-reported quality gap has not yet been independently priced. Teams running open models at real volume should also weigh serving quality: the same weights behave differently across hosts depending on quantization, speculators, and caching, which is where Morph's codegen-tuned serving earns its keep on the models it runs.

FAQ

What is Kimi K3?

Moonshot AI's flagship MoE model, launched July 16, 2026: 2.8T total parameters, 1M-token context, native vision, built on Kimi Delta Attention and Attention Residuals. Available as kimi-k3 via API and as K3 Max / K3 Cluster Max in the Kimi app.

How much does the Kimi K3 API cost?

$3/M input tokens, $0.30/M on cache hits, $15/M output tokens, flat at any context length. A 10-30% top-up bonus promotion runs through August 11, 2026.

Is Kimi K3 open source?

Not yet. No weights, license, or model card as of launch day. The K2 family's Modified MIT precedent makes an eventual open release likely but unconfirmed.

Is Kimi K3 better than GLM-5.2?

On vendor-reported agentic benchmarks, yes: 88.3 vs 81.0 on Terminal-Bench (2.0 vs 2.1) and stronger browse/search numbers. On price, no: roughly 3x the per-token cost. On verifiability, no: GLM-5.2's weights are on Hugging Face under MIT and its scores have independent replication; K3's do not yet.

Can I run Kimi K3 in Claude Code or Cline?

Any OpenAI-compatible client works against https://api.moonshot.ai/v1 with model kimi-k3, which covers Cline and similar tools directly. Keep the always-on reasoning in mind: agent loops that fire many small calls will pay max-effort reasoning on each one.

What happened to the 2.5T parameter figure?

That was the pre-launch leak number. The shipped spec in Moonshot's platform documentation says 2.8 trillion.