Moonshot AILanguage model
Kimi K2.7 Code Highspeed
Kimi K2.7 Code Highspeed has no overall index value yet because there are too few independent benchmark results.
PopularityRank 658 of 785
24out of 100-1 vs. last month
based on two data groups, mostly search
Summary
Kimi K2.7 Code Highspeed is described in Kimi’s official catalog as the fast serving variant of Kimi K2.7 Code. The catalog gives the coding family a 256,000-token context window, about 180 tokens per second, and up to 260 tokens per second in short contexts. Publisher sources identify no separate Highspeed model card or weight release.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 12 published benchmark results
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Kimi Code Bench v2 (Show measurement, test conditions, and source)
- Source value
- 62
- Score
- 62
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; official Kimi release notes explicitly identify K2.7 Code HighSpeed as the same model at roughly 5–6× output speed. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
Program Bench (Show measurement, test conditions, and source)
- Source value
- 53.6
- Score
- 53.6
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; HighSpeed is the same model per official release notes. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
MLS Bench Lite (Show measurement, test conditions, and source)
- Source value
- 35.1
- Score
- 35.1
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; HighSpeed is the same model per official release notes. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
Kimi Claw 24/7 Bench (Show measurement, test conditions, and source)
- Source value
- 46.9
- Score
- 46.9
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- agent
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; HighSpeed is the same model per official release notes. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
MCP Atlas (Show measurement, test conditions, and source)
- Source value
- 76
- Score
- 76
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- tool-use
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; HighSpeed is the same model per official release notes. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
MCP Mark Verified (Show measurement, test conditions, and source)
- Source value
- 81.1
- Score
- 81.1
- Metric
- score
- Unit
- No unit provided
- Benchmark version
- Kimi K2.7 Code
- Category
- tool-use
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- MoonshotAI Kimi K2.7 Code official model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Kimi K2.7 Code author-card row; HighSpeed is the same model per official release notes. Card says thinking enabled via Kimi Code CLI, temperature 1.0, top-p 0.95, 262,144-token context. Benchmark prompt, sampling/decoding settings, reasoning effort, quantization, and task run date are not stated in the cited result; none are inferred.
- Context
- The source presents a raw score without an explicit unit or scale; the published value is retained without percentage conversion.
SWE-Bench Pro (pass@1) (Show measurement, test conditions, and source)
- Source value
- 60.1
- Score
- 60.1
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
DeepSWE (pass@1) (Show measurement, test conditions, and source)
- Source value
- 31
- Score
- 31
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
Terminal-Bench 2.1 (pass@1) (Show measurement, test conditions, and source)
- Source value
- 67.4
- Score
- 67.4
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
Terminal-Bench 3 (pass@1) (Show measurement, test conditions, and source)
- Source value
- 1.4
- Score
- 1.4
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
Terminal-Bench 4 (pass@1) (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
SWE-Marathon (pass@1) (Show measurement, test conditions, and source)
- Source value
- 5
- Score
- 5
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1)
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper evaluates Kimi K2.7 Code and gives each benchmark as a before-to-after pair; this row preserves the first, pre-adapter value before one epoch of GSPO on a rank-32 LoRA adapter. The abstract says the six external benchmarks span three agent harnesses but does not map a harness to each benchmark. Kimi's official Code documentation says HighSpeed is built on the same model with identical coding ability; the paper does not report a separate HighSpeed run.
- Context
- The captured abstract does not state a numeric unit, per-benchmark sample size, confidence interval, benchmark revision, or harness mapping; the full paper was not opened. The first value is the paper's pre-adapter result, not a new HighSpeed-specific run. No percent unit or per-task sample count is inferred.
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Kimi K2.7 Code HighspeedClaude Mythos 5 24Amazon Nova Premier 24Jamba2 Mini 23Qwen 3.7 Plus 23
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Kimi K2.7 Code Highspeed, chosen by model line, developer, size, price and release date.
- Kimi K2.6Moonshot AI · same developer
- Kimi K2.7 CodeMoonshot AI · same developer
- Kimi K2.5Moonshot AI · same developer
- Kimi K2 ThinkingMoonshot AI · same developer
- Kimi K2 0905Moonshot AI · same developer
- Kimi K2 0711Moonshot AI · same developer
- Kimi K2 Thinking TurboMoonshot AI · same developer
- Kimi K2Moonshot AI · same developer
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Further details
Prices
At Moonshot AI from $39 a month (Kimi Pro). Through other providers from $1 a month (Command Code Go, with limited credits). 8 plans in Germany.
| Plan | Monthly list price | Note |
|---|---|---|
| Kimi ProMoonshot AI | $39 | developer's own plan |
| Kimi MaxMoonshot AI | $99 | developer's own plan |
| Kimi UltraMoonshot AI | $199 | developer's own plan |
| Command Code Go | $1 | with limited credit |
| Command Code GOAT | $10 | with limited credit |
| Command Code Pro | $20 | with limited credit |
| Command Code Max 10× | $100 | with limited credit |
| Command Code Max 20× | $200 | with limited credit |
Frequently asked questions about Kimi K2.7 Code Highspeed
What is the knowledge cutoff of Kimi K2.7 Code Highspeed?
January 2025, according to Moonshot AI. That is 17 months before release.
Stated by Moonshot AI
JanAprJulOctJanAprKnowledge cutoffJanuary 2025ReleasedJun 12, 2026
How large is the context window of Kimi K2.7 Code Highspeed?
262,144 tokens. That is how much text Kimi K2.7 Code Highspeed reads at once, about 390 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Kimi K2.7 Code HighspeedRank 67 of 103262K
Kimi K2.6262K
Qwen 3.8 27B262K
Kimi K2.7 Code262K
Kimi K2.5262K
Nemotron 3 Ultra262K
Kimi K2 Thinking262K
Can I run Kimi K2.7 Code Highspeed locally?
Yes, the weights are open. With 1T parameters it needs a lot of memory, more a server than a laptop.
Stated by Moonshot AI
- Open weights
- yes
- License
- Modified MIT
- Parameters
- 1T
- active per token
- 32B
Where can I use Kimi K2.7 Code Highspeed?
At Moonshot AI with a subscription from $39 a month. Including other providers that makes 8 plans in Germany.
Subscription comparison, prices in Germany incl. VAT
At Moonshot AI
3 plans, from $39 a month (Kimi Pro).
Through other providers
5 plans from 1 provider, from $1 a month (Command Code Go), with limited credits.
Sources and data date
Show all 13 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 27, 2026. |
| Output | Documented modalities. Retrieved October 27, 2026. |
| Research status | Research date October 15, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | benchmark-units-and-task-dates Note Moonshot publishes model-card values and a common Kimi Code CLI evaluation configuration, but does not state the units or per-benchmark task dates. |
| Unresolved | per-benchmark evaluation details Note The captured abstract does not state units, per-benchmark N, confidence intervals, benchmark revisions, or which harness ran each task. The paper was not opened beyond its abstract and does not report a separate HighSpeed run. |
| Additional source | Hugging Face (retrieved October 15, 2026) |
| Additional source | Cross-Benchmark Transfer from RL on Agentic Coding Tasks (arXiv:2610.00890v1) (retrieved October 19, 2026) |
| Additional source | Model metadata (retrieved October 15, 2026) |
| Additional source | Kimi API model list · Editorial description reviewed October 2, 2026 |
| Additional source | Kimi K2.7 Code resource · Editorial description reviewed October 2, 2026 |
| Additional source | Kimi K2.7 Code model card · Editorial description reviewed October 2, 2026 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |