Kimi K3 (Moonshot AI) vs DeepSeek V4 Pro 0813 (DeepSeek)
DeepSeek V4 Pro 0813 (DeepSeek) is 85% cheaper than Kimi K3 (Moonshot AI) ($2.64 vs $18.00 blended per 1M tokens).
Kimi K3 (Moonshot AI) on Moonshot AI vs DeepSeek V4 Pro 0813 (DeepSeek) on DeepSeek — open weight flagships. Every figure in the table below is the value shown on each model's own endpoint page, with its source and verification date. What the makers declare about the models themselves is further down.
| Metric | Kimi K3 (Moonshot AI) | DeepSeek V4 Pro 0813 (DeepSeek) |
|---|---|---|
| Input $/1M | $3 | $0.66 |
| Cached input $/1M | $0.3 | $0.022 |
| Output $/1M | $15 | $1.98 |
| Blended $/1M | $18 | $2.64 |
| Context window served | 1M | 1M |
| Max output served | — | 393.2K |
| Intelligence (Epoch) | 158 | 156 |
Highlighted cell = better on that row (lower price/latency, higher context/intelligence/throughput). "—" = not measured, or (cached input) no cache pricing published by that provider.
Independent benchmarks
Scores read on the public leaderboards on 2026-09-19 — no maker-reported numbers here (those live in the spec sheets below: a vendor's own harness, scaffold and pass@k make self-reported scores incomparable across vendors).
| Benchmark | Kimi K3 (Moonshot AI) | DeepSeek V4 Pro 0813 (DeepSeek) |
|---|---|---|
| LMArena text (Elo) | 1485 ±5 | 1458 ±4 |
| LMArena code (Elo) | 1674 ±11 | 1580 ±11 |
LMArena Elo comes from one arena with one voter pool, so the two cells compare directly. Sources: tbench.ai (Apache-2.0) · LMArena (CC-BY-4.0).
What it costs at your volume
| Monthly workload | Kimi K3 (Moonshot AI) | DeepSeek V4 Pro 0813 (DeepSeek) |
|---|---|---|
| Side project · 5M in / 1M out | $30 | $5.28 |
| Production app · 50M in / 10M out | $300 | $52.8 |
| Heavy pipeline · 500M in / 100M out | $3,000 | $528 |
Straight token math from the verified prices above — no caching, no batch discounts. If your workload reuses prompts, cached input ($0.3 vs $0.022/1M) can shift this table substantially. Model your own traffic in the LLM cost calculator.
Neither price has moved since we started tracking them (2026-09-03). Full price history on each endpoint page.
Which to pick
Kimi K3 (Moonshot AI) if…
- code is the workload — LMArena code Elo 1674 vs 1580
- general answer quality — LMArena text Elo 1485 vs 1458
DeepSeek V4 Pro 0813 (DeepSeek) if…
- cost per token is the constraint — $2.64 vs $18 blended per 1M
Every line above is derived from a measured row on this page — nothing editorial. What the table can't weigh: rate limits, SDK ecosystem, data-retention terms — those are on each endpoint page.
Same weights, other hosts
The table at the top compares one endpoint per side. But the same model is often sold by several hosts — open weights re-hosted, or a closed model resold through clouds — at different prices, sometimes a bigger spread than the head-to-head itself.
kimi-k3 is sold by 17 hosts we track. Cheapest: Chutes at $11.35/1M blended. All 17 priced side by side on the kimi-k3 page.
deepseek-v4 is sold by 23 hosts we track. Cheapest: Relace at $2.5/1M blended. All 23 priced side by side on the deepseek-v4 page.
What the makers declare
The table above compares two endpoints — what each host charges, how much of the window it serves, how fast it answers. This is the other register: what the maker declares about the weights, the same whichever host you buy from. Where a row here carries a different number from the same row above, both are true — one describes the model, the other describes the endpoint that serves it.
kimi-k3maker-declared
- Context window
- 1,048,576 tokens
- Parameters
- 2,779,931,837,184≈ 2.78T
- Active experts
- 16 of 896 active per token
- Weight precision
- not published by the maker
- License
- Modified MIT — open weights, conditionalcuratedScope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ April 2026 third-party estimate — not the maker's figurecurated
- Modalities
- text · image
- Max output (official)
- 1,048,576 tokensmaker's docs · read 2026-09-19
- Release date
- not published by the makermaker's docs · read 2026-09-19
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- KimiK3ForConditionalGeneration
- Layers
- 93
- Vocabulary
- 163,840 tokens
- Weights first published
- June 13, 2026
- Weights last updated
- September 2, 2026
moonshotai/Kimi-K3 · config.json @ f831ab6 · read 2026-10-08 · 6 of 10 fields published by the maker
deepseek-v4maker-declared
- Context window
- 1,048,576 tokens
- Parameters
- 1,650,497,936,906≈ 1.65T
- Active experts
- 6 of 384 active per token
- Weight precision
- fp8
- License
- MIT — open weights, permissivecurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ March 2026 third-party estimate — not the maker's figurecurated
- Modalities
- text
- Max output (official)
- to verify
- Release date
- to verify≈ August 13, 2026 third-party catalogue — not the maker's figure
- Maker lifecycle status
- Deprecated — retirement announced for September 14, 2026, still served until then · superseded by DeepSeek V4.1 Flashmaker's docs · read 2026-09-10
Also declared in the weights
- Architecture class
- DeepseekV4ForCausalLM
- Layers
- 61
- Vocabulary
- 129,280 tokens
- Weights first published
- August 13, 2026
- Weights last updated
- August 13, 2026
- Paper
- arXiv:2606.19348
deepseek-ai/DeepSeek-V4-Pro-0813 · config.json @ 72e1d32 · read 2026-10-21 · 7 of 10 fields published by the maker
A head-to-head between two models is the narrow question. The wider one — the same weights sold at different prices by several hosts — is in the catalogue, one line per model with every host's price.
Sources: Kimi K3 (Moonshot AI) endpoint · DeepSeek V4 Pro 0813 (DeepSeek) endpoint · Kimi K3 (Moonshot AI) price verified 2026-10-23 · DeepSeek V4 Pro 0813 (DeepSeek) price verified 2026-10-22