Kimi K2

Moonshot's open-weights agentic model: 262K context, reasoning, tool calling and structured output.

name
Kimi K2
vendor
Moonshot AI
model_id
kimi-k2
context_window
262K
max_output
262K
input_per_mtok
see provider
output_per_mtok
see provider
strengths
Moonshot's open-weights agentic model: 262K context, reasoning, tool calling and structured output.
provider
Moonshot AI
family
Kimi K2
release_date
2026-04-21 source
last_updated
2026-06-15
open_weights
true source
license
open (see Hugging Face) source
params_total
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — total parameter count not published for this model.
params_active
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — active (MoE) parameter count not published for this model.
tool_call
true source
reasoning
always source
structured_output
true source
attachment
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — accepted attachment/file modalities not enumerated on the fetched entry; confirm at build.
temperature
true source
knowledge_cutoff
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — knowledge cutoff not published on the fetched entry; confirm at build.
context_advertised
262K source
context_effective
— verify-against-primary-at-build ↗ No published measured long-context recall benchmark for this model; do not derive effective context from advertised.
price_input
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — canonical input price not reliably resolvable (placeholder pricing on the entry); see provider.
price_output
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — canonical output price not reliably resolvable (placeholder pricing on the entry); see provider.
price_cache_read
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — cache-read rate not captured numerically from a primary source in this pass.
price_cache_write
— verify-against-primary-at-build ↗ https://models.dev/models/moonshotai/kimi-k2/ — cache-write rate not captured numerically from a primary source in this pass.
cost_per_full_window
— verify-against-primary-at-build ↗ Derived at build from price_input x advertised context once confirmed.
cost_per_agent_task
— verify-against-primary-at-build ↗ Derived at build from the cache-aware agent-task model once cache rates are confirmed.
modalities
text-in,text-out source
gpqa_diamond
— verify-against-primary-at-build ↗ https://artificialanalysis.ai/ — GPQA-Diamond for this model to confirm at build.
swe_bench_verified
— verify-against-primary-at-build ↗ https://www.swebench.com/ — SWE-bench Verified score for this model to confirm at build.
terminal_bench
— verify-against-primary-at-build ↗ https://www.tbench.ai/ — Terminal-Bench score for this model to confirm at build.
tau2_bench
— verify-against-primary-at-build ↗ Primary tau2-Bench leaderboard — score for this model to confirm at build.
bfcl_tool_use
— verify-against-primary-at-build ↗ https://gorilla.cs.berkeley.edu/leaderboard.html — BFCL tool-use score for this model to confirm at build.
aa_index
— verify-against-primary-at-build ↗ https://artificialanalysis.ai/ — Intelligence Index for this model to confirm at build.
lmarena_elo
— verify-against-primary-at-build ↗ https://lmarena.ai/leaderboard — human-preference Elo for this model to confirm at build.
tokens_per_sec
— verify-against-primary-at-build ↗ https://artificialanalysis.ai/ — measured throughput for this model to confirm at build.
ttft
— verify-against-primary-at-build ↗ https://artificialanalysis.ai/ — measured time-to-first-token for this model to confirm at build.
hallucination_rate
— verify-against-primary-at-build ↗ No primary hallucination benchmark captured for this model in this pass.
agent_readiness_score
— verify-against-primary-at-build ↗ Score withheld: R_tool (BFCL/tau2-Bench), R_ctx (effective context), R_cost (cache rates) and R_latency inputs are not yet sourced. Compute per /models/agent-readiness-score once inputs confirmed.
score_confidence
partial
source_url
https://models.dev/models/moonshotai/kimi-k2/
source_type
models.dev
last_verified
2026-06-15
_provenance_note
context_advertised and max_output both 262,144 per models.dev. Canonical entry showed $0.00/$0.00 (placeholder) -> price deferred to 'see provider' rather than asserting a free price.

← all The Frontier Model Matrix · .md · JSON