Kimi K3
Sold by 29 sellers, timed at 110 tok/s, scored 17.4 on the Intelligence Index, on 20 Design Arena boards, and with results on 2 openly licensed boards.
retrieved sha256 · 0bb53a77713f
The same weights served by 6 providers, each at its own list price. OpenRouter's rate is its routed rate across whichever of them it uses, and it need not match any one of them.
What sellers charge
29 sellers · 29.41x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.153 at Wafer to $4.50 at Parasail. 2 sellers undercut the $0.192 listed; 26 are dearer. A faint tick declares no number format. Every seller is named on the endpoint ledger below.
The same sellers grouped by the number format each declares, on the same scale. The fp4 group alone spans 95% of the range, so the cheap end is not cheap for its number format. A declaration is checked by nobody: this groups the rows and certifies none.
Measured speed
19,093 requests · 28 endpoints · 30-minute windowOpenRouter's own timing of its routed traffic, weighted by requests, over one 30-minute window read . Every route is on the endpoint ledger below.
Every endpoint
29 priced · 28 timed · cheapest first- Waferwafer/fast$0.153$16.00$0.152mxfp4944knot matchednonenonenot matched100%
- Relacerelace/fp4$0.1805$11.0055$0.18fp4944k90 tok/s46.31.29s638 req99.85%
- Waferwafer · listed$0.192$16.00$0.182mxfp4944knot matchednonenonenot matched100%
- InferenceNetinference-net/fp4$0.245$15.75$0.214fp4944knot matchednonenonenot matched99.75%
- InferenceNetinference-net/us$0.245$15.75$0.214fp4944knot matchednonenonenot matched99.76%
- Waferwafer/us$0.276$9.60$0.13mxfp4944knot matchednonenonenot matched99.84%
- Sail Researchsail-research/fp4$0.345$16.58$0.1725fp4944knot timednonenonenone0.00%
- InferenceNetinference-net/fast$0.45$18.00$0.363fp4944knot matchednonenonenot matched100%
- InferenceNetinference-net/us/fast$0.45$18.00$0.363fp4944knot matchednonenonenot matched100%
- KTkt$0.45$15.00$0.30not declared944k102 tok/s371.37s483 req100%
- Io Netio-net/fp4$0.90$14.00$0.15fp4131k101 tok/s52.11.54s345 req99.90%
- Makoramakora/fp4$1.00$18.00$0.50fp4944k98 tok/s58.61.17s166 req100%
- Chuteschutes/mxfp4$1.85$9.50$0.185mxfp466k24 tok/s19.52.81s69 req100%
- DigitalOceandigitalocean$2.55$12.95$0.255not declared944k50 tok/s31.61.52s175 req100%
- DeepInfradeepinfra/mxfp4$2.85$14.25$0.285mxfp4944k94 tok/s58892ms828 req100%
- Alibabaalibaba$3.00$15.00$0.30not declared900k53 tok/s34.41.68s119 req100%
- Amazon Bedrockamazon-bedrock/us-east-2$3.00$15.00$0.30not declared128knot matchednonenonenot matched99.13%
- BaseTenbaseten/fp8$3.00$15.00$0.30fp8262k146 tok/s96.4540ms169 req100%
- Fireworksfireworks$3.00$15.00$0.30not declared944knot matchednonenonenot matched100%
- Modalmodal/mxfp4$3.00$15.00$0.30mxfp4944k97.5 tok/s46604ms65 req100%
- Moonshot AImoonshotai$3.00$15.00$0.30not declared944k24 tok/s142.75s984 req100%
- Nebiusnebius/fp4$3.00$15.00$0.30fp4944k134 tok/s63651ms772 req100%
- Parasailparasail/fp4$3.00$15.00$0.30fp4944knot matchednonenonenot matched100%
- Riverriver$3.00$15.00$0.30not declared1.05M154 tok/s103268ms1,302 req99.85%
- Togethertogether$3.00$15.00$0.30not declared944k59 tok/s391.04s1,948 req99.76%
- Amazon Bedrockamazon-bedrock/us-east-1$3.30$16.50$0.33not declared128knot matchednonenonenot matched92.68%
- Fireworksfireworks/fast$4.50$22.50$0.45not declared944knot matchednonenonenot matched99.58%
- Fireworksfireworks/us$4.50$22.50$0.45not declared944knot matchednonenonenot matched100%
- Parasailparasail/fast$4.50$22.50$0.45fp4944knot matchednonenonenot matched100%
- Amazon Bedrocknot matchednot matchednot matchednot matchedno offer87 tok/s44562ms32 reqno offer
- Amazon Bedrocknot matchednot matchednot matchednot matchedno offer59 tok/s392.88s288 reqno offer
- Fireworksnot matchednot matchednot matchednot matchedno offer94 tok/s46.5936ms118 reqno offer
- Fireworksnot matchednot matchednot matchednot matchedno offer43 tok/s27.81.51s120 reqno offer
- Fireworksnot matchednot matchednot matchednot matchedno offer47 tok/s271.65s472 reqno offer
- InferenceNetnot matchednot matchednot matchednot matchedno offer130 tok/s59.4695ms290 reqno offer
- InferenceNetnot matchednot matchednot matchednot matchedno offer132 tok/s69.2461ms415 reqno offer
- InferenceNetnot matchednot matchednot matchednot matchedno offer139 tok/s51.1530ms588 reqno offer
- InferenceNetnot matchednot matchednot matchednot matchedno offer141 tok/s75555ms729 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer50 tok/s22907ms522 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer98.5 tok/s61.1405ms97 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer142 tok/s91604ms1,134 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer135 tok/s90469ms3,754 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer132 tok/s78444ms2,471 reqno offer
Each track runs from a route's p90, the rate nine in ten of its requests met or beat, up to its median. Weighted by requests, the figure above reports where the traffic went rather than what one route does: 18 of the 28 timed routes are slower than it, and the middle route runs at 97.8 tok/s. The widest median-to-p90 gap is KT's, 2.74x.
Amazon Bedrock publishes 2 offers and had 2 routes timed; Fireworks publishes 3 offers and had 3 routes timed; InferenceNet publishes 4 offers and had 4 routes timed; Parasail publishes 2 offers and had 2 routes timed; Wafer publishes 3 offers and had 3 routes timed. The two documents share no route id, so those offers and routes are printed apart rather than paired.
Uptime is OpenRouter's own figure for each seller over the trailing day, read once a day, and the column prints the latest reading, taken . 24 of the 37 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: Fireworks (fireworks/spot) at 0.00% on 2 Nov 2026, Sail Research (sail-research/fp4) at 0.00% on 21 Nov 2026, Fireworks (fireworks/fast) at 43.04% on 13 Nov 2026, Fireworks (fireworks/us) at 62.50% on 25 Oct 2026, and 20 more. Source: OpenRouter's public endpoints API, one reading a day for each seller: 1,211 sellers carried one on the latest of the 23 days held.
Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. List prices, not what anybody paid; a discount is the seller's own. Read . Speed is OpenRouter's own timing over one 30-minute window, read .
What each seller accepts
20 parameters · 29 sellers- include_reasoningevery seller
- max_tokensevery seller
- reasoningevery seller
- reasoning_effortevery seller
- temperature27 of 29not Amazon Bedrock (amazon-bedrock/us-east-1) and Amazon Bedrock (amazon-bedrock/us-east-2)
- top_p27 of 29not Amazon Bedrock (amazon-bedrock/us-east-1) and Amazon Bedrock (amazon-bedrock/us-east-2)
- stop26 of 29not Alibaba, DigitalOcean, and KT
- tools26 of 29not Chutes, Parasail (parasail/fast), and River
- response_format25 of 29
- structured_outputs25 of 29
- presence_penalty24 of 29
- frequency_penalty23 of 29
- tool_choice23 of 29
- top_k23 of 29
- repetition_penalty19 of 29
- seed19 of 29
- logprobs17 of 29
- top_logprobs17 of 29
- logit_bias15 of 29
- min_p9 of 29
4 of these 20 are accepted by every one of the 29 sellers that publish a list; the other 16 are declared by some sellers and not others, so what a request sending one of them gets depends on which seller serves it. Each list is the seller's own declaration in OpenRouter's endpoints API, read , and nothing here tested one.
Price position
342 priced modelsBlended $4.144 per million tokens, 3 in to 1 out: cheaper than 54 of the 342 priced models in the catalog. Each bar is one stretch of the catalog on a logarithmic price scale; the mark is this listing.
Measured quality
145 models with an index · 132 on a design arena boardEach mark is one board this model stands on, placed at the share of that board it scores above, and led to its number in the list under the rule. The columns and the ledger below carry the figures.
- Agentic game development · Agents1201 Elowin rate 54.0% · source rank 4the board, 25 ranked
- Android native · Agents1255 Elowin rate 57.1% · source rank 7the board, 41 ranked
- blender3d · Agents1124 Elowin rate 40.4% · source rank 9the board, 25 ranked
- codetovideo · Agents1294 Elowin rate 65.3% · source rank 4the board, 26 ranked
- Full stack · Agents1289 Elowin rate 61.4% · source rank 5the board, 44 ranked
- Godot game development · Agents1193 Elowin rate 48.6% · source rank 13the board, 34 ranked
- HTML slides · Agents1152 Elowin rate 55.6% · source rank 5the board, 25 ranked
- Mobile apps · Agents1252 Elowin rate 56.5% · source rank 10the board, 49 ranked
- pdf · Agents1279 Elowin rate 65.1% · source rank 5the board, 36 ranked
- python-pptx slides · Agents1223 Elowin rate 53.9% · source rank 9the board, 29 ranked
- spreadsheet · Agents1052 Elowin rate 27.7% · source rank 19the board, 32 ranked
- Web apps · Agents1277 Elowin rate 58.0% · source rank 9the board, 47 ranked
- 3D scenes · Models1365 Elowin rate 63.7% · source rank 11the board, 113 ranked
- ASCII art · Models1351 Elowin rate 68.3% · source rank 2the board, 67 ranked
- Code · Models1351 Elowin rate 60.3% · source rank 6the board, 120 ranked
- Data visualisation · Models1349 Elowin rate 63.5% · source rank 4the board, 110 ranked
- Game development · Models1354 Elowin rate 55.2% · source rank 10the board, 120 ranked
- SVG · Models1296 Elowin rate 59.5% · source rank 9the board, 83 ranked
- UI components · Models1340 Elowin rate 60.6% · source rank 10the board, 116 ranked
- Websites · Models1337 Elowin rate 59.5% · source rank 4the board, 126 ranked
Position 26 of the 258 models Epoch has placed, which is not the population of either board above, on a scale anchored at Claude 3.5 Sonnet 130 and GPT-5 150. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
Benchmark results
2 of 6 boards · 40 results| Task | All runs | Effort | Score |
|---|---|---|---|
| Chess Puzzles | 167 | max | 0.36 |
| high | 0.37 | ||
| low | 0.18 | ||
| FrontierMath-Erdos | 16 | max | 0 |
| FrontierMath-Tier-4-v2-Private | 84 | max | 0.39 |
| FrontierMath-Tiers-1-3-v2-Private | 128 | max | 0.72 |
| Furniture Assembly | 42 | max | 0.34 |
| GPQA diamond | 342 | max | 0.93 |
| high | 0.92 | ||
| low | 0.85 | ||
| MirrorCode | 48 | max | 0.54 |
| Mystery Game Puzzles | 149 | max | 0.26 |
| low | 0.12 | ||
| OTIS Mock AIME 2024-2025 | 319 | max | 0.97 |
| high | 0.93 | ||
| low | 0.69 | ||
| SimpleQA Verified | 106 | max | 0.51 |
Mean accuracy, 0 to 1, on the task the row names, as Epoch AI runs it and averaged over repeated runs. All eighteen of their tasks, including the held-out FrontierMath tiers, named as they name them. The model id carries the effort setting the run used. Epoch AI Benchmarking Hub names these runs kimi-k3_max, kimi-k3_high and kimi-k3_low, and the effort beside each score says which run it is. A run missing from a task's rows has no published result there, which is not a score of zero. All runs is how many runs the publisher scored on the task, every model's. Read from epoch.ai ↗, which dates its own data . Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
| Task | Score |
|---|---|
| AMPS_Hard | 97 |
| code_completion | 82.61 |
| code_generation | 80.28 |
| connections | 100 |
| consecutive_events | 89.76 |
| integrals_with_game | 54 |
| javascript | 68.18 |
| logic_with_navigation | 80 |
| math_comp | 95.10 |
| olympiad | 91.65 |
| paraphrase | 74.37 |
| plot_unscrambling | 72.58 |
| python | 65 |
| simplify | 66.37 |
| spatial | 100 |
| story_generation | 75.37 |
| summarize | 69.35 |
| tablejoin | 48.40 |
| tablereformat | 98.04 |
| theory_of_mind | 82.69 |
| typescript | 53.33 |
| typos | 84 |
| zebra_puzzle | 100 |
Percent correct on the task the row names, higher is better. All 23 published columns; there is no average to carry, because the site computes its headline in the browser from a weighting this does not reimplement. LiveBench names this run kimi-k3, with no effort setting written on it. Read from livebench.ai ↗, which dates its own data . LiveBench, Apache-2.0. Results produced by the LiveBench maintainers on questions refreshed monthly.
2 of the 6 openly licensed boards Vitna mirrors carry a result for Kimi K3, each printed as its publisher published it and none added to another. MTEB, HELM, Vectara hallucination leaderboard and Aider polyglot have not published one under a name that matches its listing, and names are matched to the catalog exactly and never guessed. A task that is a link opens its column on /capability, with every run scored on it; an effort that is a link opens its run's row there, where the run is ranked on Vitna's scale. Every board, with its licence and what its publisher measures, is on /benchmarks.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Source: each model's own Hugging Face repository (huggingface.co), as of 20 Nov 2026. 162 repositories read, 4 did not answer.
Licence, parameter count, model class and gating: each model's own Hugging Face repository, the one OpenRouter names for the listing.
A licence is the tag the uploader put on the repository, read from nothing else and verified by nobody.
A parameter count is the total the repository's safetensors files declare; for a mixture of experts that is every expert, not the share one token runs through.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Vitna ran none of these evaluations and adjusted none of these figures.
