Quick verdict: test Kimi K3 now if you need Kimi’s newest long-horizon, multimodal, 1M-context model. Do not treat its open-weights story as complete until Moonshot publishes the promised weights, license, checkpoint details, and serving instructions by July 27, 2026.

K3 is a serious frontier-adjacent Kimi upgrade: Moonshot describes a 2.8T-parameter model with native vision, Kimi Delta Attention, Attention Residuals, 1M context, and strong launch-suite results. The current AIHackers position is narrower: K3 is worth a controlled eval, but the launch evidence is mixed by provenance. Moonshot’s benchmark table is vendor evidence, Artificial Analysis is an early independent signal, and AIHackers has not yet run a site-owned repository test.

Quick Facts

SpecKimi K3
ProviderMoonshot AI / Kimi
Model IDkimi-k3
Parameters2.8T total parameters, vendor-stated
MoE routing16 of 896 experts active, vendor-stated
Architecture notesKimi Delta Attention, Attention Residuals, Stable LatentMoE
Context window1M tokens
Input / outputText and image input; text output in current independent listing
API reasoningAlways-on thinking; pay-as-you-go API currently supports reasoning_effort: max only
Kimi Code reasoningMembership model ID k3 supports low, high, and max; access and 256K/1M context depend on plan tier
API pricing$0.30 cache-hit input / $3.00 cache-miss input / $15.00 output per 1M tokens
Quantization notesMoonshot says quantization-aware training uses MXFP4 weights and MXFP8 activations from SFT onward
Deployment noteMoonshot recommends supernode deployments with 64 or more accelerators and says vLLM KDA support is planned alongside the model
WeightsPromised by July 27, 2026; not inspectable at this review

Kimi’s current platform model list also says kimi-k2.5 and the moonshot-v1 series are closed to new registered users after K3’s release, with full platform sunset on August 31, 2026. Treat that as a migration clock for older Kimi integrations, not as proof that K3 is the cheapest replacement.

Benchmark Evidence

SourceKimi K3 signalHow to read it
Moonshot launch blogK3 max reasoning is shown near GPT-5.6 Sol, Opus 4.8, GLM-5.2, and Fable 5 across coding, agentic, reasoning, and vision tablesVendor-reported launch suite; benchmark harnesses and fallback behavior vary
Artificial AnalysisIntelligence Index 57, 1M context, 62 output tokens/s, $3 input / $15 output, proprietary statusEarly independent shortlist signal; not an AIHackers repo eval
AIHackersNot verifiedNo accepted-task field test has been run yet

benchmark artifact

Kimi K3 Evidence Snapshot

ModelProviderStatusContextInput priceOutput priceCoding signalTool-use signalBenchmark evidenceSpeedVerdictSourcesChecked
Kimi K3Moonshot AIactive
Active Kimi API/product flagship; full weights promised by July 27, 2026 and not inspectable at this review.
1M$3.00 / 1M$15.00 / 1MMoonshot reports strong max-reasoning launch-suite coding and agent results; AIHackers repo eval is not verified.Kimi API and Kimi Code support K3; Kimi docs warn to preserve full assistant history and avoid mid-session model switching.
  • Artificial Analysis Intelligence Index v4.1: 57 (independent)
  • Artificial Analysis output speed: 62 tokens/s (independent)
  • Moonshot launch benchmark suite: vendor-reported max-reasoning table (vendor)
  • AIHackers repo eval: not verified (site-owned)
Artificial Analysis reports 62 output tokens/s and flags high verbosity; measure total output cost per accepted task.Test as Kimi's newest 1M-context frontier-adjacent lane; keep K2.7 Code for cheaper routine Kimi coding until K3 passes local CAR tests.Kimi K3 launch blog [archive], Kimi K3 quickstart [archive], Kimi K3 API pricing [archive], Kimi current model list [archive], Kimi Code model configuration, Artificial Analysis: Kimi K3 [archive]2026-07-18
GLM-5.2Z.AIactive
Current Z.AI flagship coding model and supported-tool value lane.
1M$1.40 / 1M$4.40 / 1MZ.AI reports 62.1 on SWE-Bench Pro and 81.0 on Terminal-Bench 2.1.Supported-tool coding lane; BFCL score not imported.
  • Artificial Analysis Intelligence Index v4.1: 51 (independent)
  • SWE-Bench Pro: 62.1 (vendor)
  • Terminal-Bench 2.1: 81.0 (vendor)
Artificial Analysis flags higher output-token use; measure total cost per successful task.July value pick to test for supported coding-tool workflows; keep Opus/GPT for final arbitration until local evals pass.Z.AI GLM-5.2 overview [archive], Z.AI pricing [archive], Artificial Analysis: GLM-5.2 article [archive], Artificial Analysis Intelligence Index v4.1, SWE-bench, Berkeley Function Calling Leaderboard2026-06-28
Kimi K2.7 CodeMoonshot AIactive
Cheaper routine Kimi coding API lane; HighSpeed is the same model at higher token prices.
256K$0.95 / 1M$4.00 / 1MKimi K2.7 Code remains the lower-cost 256K coding lane after K3; independent normalized benchmarks are not imported.OpenAI-compatible API; thinking mode required in the documented K2.7 Code quickstart.
  • Program-Bench improvement vs K2.6: +10.4% (vendor)
  • MCP Mark Verified improvement vs K2.6: +11.4% (vendor)
  • SWE Marathon improvement vs K2.6: +76.2% (vendor)
  • Reasoning-token use vs K2.6: 30% lower (vendor)
  • AIHackers repo eval: not verified (site-owned)
HighSpeed model ID exists at a higher token price; latency not independently measured here.Cheaper routine Kimi coding API lane when Kimi routing fits and 256K context is enough.Kimi K2.7 Code quickstart [archive], Kimi K2.7 Code pricing [archive], Kimi Code K2.7 release notes [archive], SWE-bench, Berkeley Function Calling Leaderboard2026-06-28
Claude Opus 4.8Anthropicactive
Current generally available Opus-tier premium baseline.
1M$5.00 / 1M$25.00 / 1MPractical premium Claude baseline; AIHackers recommends task-level comparison rather than blanket default routing.Claude API and Claude-native workflow baseline; third-party routing must follow Anthropic terms.
  • Artificial Analysis Intelligence Index v4.1: 56 (independent)
  • Artificial Analysis output speed: 57.3 tokens/s (independent)
Artificial Analysis measured 57.3 output tokens/s; provider and workload latency vary.Premium review, architecture, hard-debugging, and final-arbitration lane.Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 4.8 [archive], Artificial Analysis Intelligence Index v4.1, LMArena leaderboard dataset, Berkeley Function Calling Leaderboard2026-07-18
GPT-5.6 SolOpenAIpreview
Selected API organizations and Codex workspaces only; no public enrollment and no ChatGPT access during preview.
not published$5.00 / 1M$30.00 / 1MOpenAI reports a new state of the art on Terminal-Bench 2.1; expanded and independent results are pending.API and Codex preview; max reasoning and ultra multi-agent modes are vendor-documented.
  • Terminal-Bench 2.1: vendor reports new state of the art (vendor)
  • AIHackers repo eval: not verified (site-owned)
OpenAI announced a selected-customer Cerebras preview for July; production latency is not verified.Restricted-preview evaluation only; keep GPT-5.5 as the active OpenAI baseline.OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive]2026-06-28

Kimi K3 has vendor launch evidence and an early Artificial Analysis row. AIHackers has not yet run a controlled repository evaluation, and open weights remain pending until the promised July 27 release.

Do not convert Moonshot’s launch table—or one independent aggregate—into a blanket task-level winner claim. In the current Artificial Analysis v4.1 snapshot, K3 is 57 at 62 tokens/s and Opus 4.8 is 56 at 57.3 tokens/s. That is a shortlist signal under one benchmark mix, not proof that K3 will produce better accepted patches, reviews, or production outcomes.

Kimi K3 vs Current Alternatives

AlternativeSource-labeled comparison
GPT-5.6 SolMoonshot compares K3 against Sol in its launch suite. AIHackers still treats GPT-5.6 as restricted-preview evidence, not normal availability.
Claude Opus 4.8Opus remains the practical premium Claude baseline. K3 is cheaper on listed output price but not proven cheaper per accepted patch.
GLM-5.2GLM-5.2 keeps the low-cost 1M-context value lane at $1.40/$4.40. K3 costs more and should be tested as a stronger escalation, not a default replacement.
Kimi K2.7 CodeK2.7 Code remains the cheaper routine Kimi coding lane at 256K context. Route latest-Kimi and 1M/multimodal frontier intent to K3; route budget Kimi coding intent to K2.7 Code.

The practical switch is not just an ID swap. K3 changes context, reasoning behavior, price, cache behavior, and migration risk.

Production Caveats

  • Preserve the full assistant history. In multi-turn and tool-calling API flows, pass the complete prior assistant message back unchanged, including thinking/reasoning fields—not only the final content. Moonshot warns that dropping this history can make K3 unstable.
  • Do not switch to K3 mid-session. K3 was trained with preserved thinking history. Start a new session instead of moving an active K2.x or other-model conversation to K3.
  • Constrain proactivity explicitly. For bounded production agents, state approval boundaries, permitted actions, and stop conditions in the system prompt or AGENTS.md; Moonshot warns that K3 may otherwise improvise around ambiguity or minor obstacles.
  • Keep web_search out of production for now. The current Kimi API guide says the feature is being updated and is not recommended for near-term production workflows.

These caveats apply to the current pay-as-you-go API contract. Kimi Code membership is a separate managed product: its k3 route currently exposes low/high/max effort, while 256K versus 1M context depends on the membership tier.

CAR Cannot Be Computed Yet

AIHackers cannot compute Cost per Accepted Result (CAR) for K3 until there is an accepted-task field test.

1
CAR = (model/tool cost + human review and cleanup hours * loaded hourly rate) / accepted tasks

Run this five-task first pass before making K3 a default route:

TaskPass condition
Repository mapCorrect modules, commands, and constraints; no invented files
Bug fixMinimal patch for one real failing test
Refactor2-4 files changed while preserving local style and behavior
Code reviewFile-grounded findings on a known risky patch
Multimodal taskScreenshot or image-grounded UI/code task that uses visual evidence concretely

Record exact model ID, reasoning_effort, context size, token usage, cache hit/miss behavior, retries, latency, accepted result, and review minutes. If K3 produces long reasoning traces, the $15 output price matters more than the headline input price.

What Changes After July 27

Recheck these items when Moonshot publishes the weight release:

  • Whether the full weights are actually available from an official Moonshot-controlled location.
  • License terms, commercial-use terms, acceptable-use policy, and any geographic restrictions.
  • Checkpoint size, shard format, tokenizer, and exact architecture/config files.
  • Whether vLLM, SGLang, KTransformers, or other engines support KDA and the K3 checkpoint without vendor-only patches.
  • Realistic self-hosting requirements for the 64+ accelerator guidance.
  • Independent reproduction of headline benchmark results.
  • Whether Artificial Analysis changes K3 from proprietary to open-weights status.
  • Whether Kimi API docs add low/high reasoning effort modes or pricing changes after launch.

Until then, K3 is “active for API/product use, open weights pending” in the AIHackers benchmark data.

Sources


Last verified: July 18, 2026. Kimi K3 prices, model-list status, reasoning modes, independent scores, and open-weight availability can change quickly during the launch window.