Kimi K2.7 Code vs MiniMax M3

MiniMax M3 has been retired from our catalogue. Its figures below are the last published rates; calls to it are no longer served, while the other model in this comparison is.

vs

Which one, when

Both models take text, image and video in and return text, and both cover chat, code, reasoning and tools, so the split is price and scale: minimax-m3 runs $0.3 input and $1.2 output against $0.95 and $4 for kimi-k2.7-code, roughly 3.2x cheaper on input and 3.3x on output, with a 1000000-token context and 524288 max output versus 256000 and 32768. Pick minimax-m3 for very long inputs, long generations, or when you want to turn thinking off, since kimi-k2.7-code always reasons. Choose kimi-k2.7-code if you specifically want Moonshot's model within its 256000-token limit.

Benchmarks

MiniMax M3: the vendor has not published benchmark scores.

Above averageKimi K2.7 Code1 / 5
Kimi K2.7 Code MiniMax M3 other models measured peer average ★ no peer scored higher
MLS-Bench-Lite
35.1%
N/A
MCP-Atlas
76%
N/A

Vendor-published: Alibaba (Qwen) Alibaba (Qwen) Alibaba (Qwen) Alibaba (Qwen) Moonshot Moonshot OpenAI OpenAI Z.ai

Pricing

Kimi K2.7 Code MiniMax M3 Δ
Input / 1M tokens $0.95 $0.3 3.2×
Output / 1M tokens $4 $1.2 3.3×
Cache read / 1M tokens $0.19 $0.06 3.2×
Cache write - no separate charge -

Rates from the live catalogue at build time; each model page carries the current rate card.

Where they sit · input price per 1M tokens across all 91 chat models on this billing unit (log scale)

Kimi K2.7 Code · $0.95 MiniMax M3 · $0.3
$0.05 · Qwen3 VL Flash $30 · GPT-5.4 Pro

Capabilities

Kimi K2.7 Code MiniMax M3
Tool calling yes yes
Thinking control always on configurable
Prompt caching implicit (automatic) implicit (automatic)
Cache lifetime not published not published
Minimum cached prefix not published 512 tokens

Specs

Kimi K2.7 Code MiniMax M3
Input modalities text image video text image video
Output modalities text text
Released 2026-06 2026-06-01
Context window 256,000 1,000,000
Max output 32,768 524,288
Thinking parameter
  • thinking.type
  • thinking.keep
  • thinking.type
  • reasoning_split
Accepted values
type
  • accepts only enabled
keep
  • accepts only all
thinking.type
  • adaptive
  • disabled
reasoning_split
  • boolean
Default thinking on with Preserved Thinking on adaptive: thinking on, with the model deciding when extra reasoning helps

Specs are transcribed from each vendor’s documentation; a row a vendor does not publish is left out rather than inferred. Full sources: Kimi K2.7 Code · MiniMax M3

One prompt, both models, measured through the gateway

PROMPT Explain to a CFO, in exactly three sentences, why our LLM bill doubled after we enabled reasoning. CHECK exactly 3 sentences

Kimi K2.7 Code passed · 3 sentences

Enabling reasoning mode causes the model to generate an internal chain-of-thought before producing each answer, and that hidden reasoning text is billed like any other output. Those extra reasoning tokens typically add up to much more than the final response itself, so every API call now costs significantly more than before. Because our overall usage volume did not change, the doubled bill reflects that each query consumes roughly twice as many billable tokens.

out 259 tok (+174 thinking) latency 5.4 s

MiniMax M3 passed · 3 sentences

Reasoning models don't just answer — they first generate extensive internal "thinking" tokens to work through the problem step-by-step, and you're billed for all of those tokens even though users never see them. A query that previously cost ~200 tokens may now cost 2,000–10,000 once reasoning is enabled, multiplying per-query spend by roughly 5–50x while the visible output stayed the same size — that's the doubling. The lever is tuning: cap reasoning effort on simple queries, cache the reasoning traces, and reserve the full reasoning model for only your hardest 10–20% of requests, which typically restores 60–80% of the savings.

out 764 tok (+629 thinking) latency 3.5 s

Instruction following (exactly three sentences, so countable), audience shaping (CFO register), and the hidden-thinking billing gap the token meters below expose.

PROMPT This function is supposed to return the longest run of consecutive equal items, but callers report it is off by one on single-element inputs and misses runs that end at the last index. Fix it and explain the bug in one paragraph. CHECK fix passes tests

Kimi K2.7 Code passed · 8/8 cases

```python def longest_run(xs): if not xs: return 0 best = 1 cur = 1 for i in range(1, len(xs)): if xs[i] == xs[i - 1]: cur += 1 best = max(best, cur) else: cur = 1 return best ``` The bug is that the function counts *transitions* between equal neighbors instead of counting the items themselves, so a single-element list never enters the loop and returns `0` instead of `1`. Likewise, resetting `cur` to `0` on a mismatch forgets the current item that starts a new run, which causes runs that end at the last index to be undercounted. Initializing both `best` and `cur` to `1` and resetting `cur` to `1` fixes both issues.

out 410 tok (+222 thinking) latency 9.4 s

MiniMax M3 passed · 8/8 cases

The bug is that the loop starts at index 1 and treats `cur` as the number of *equal transitions* rather than the run length, so a run of *k* consecutive equal items is reported as `k − 1`, and a single-element list (where the loop body never executes) returns 0 instead of 1. A correct version tracks the actual run length by seeding `cur = 1` and resetting it to 1 (not 0) on a mismatch: ```python def longest_run(xs): if not xs: return 0 best = cur = 1 for i in range(1, len(xs)): if xs[i] == xs[i - 1]: cur += 1 else: cur = 1 if cur > best: best = cur return best ``` Now `[5]` returns `1`, `[1,1,1]` returns `3`, and a run like `[1,2,2,2]` that ends at the last index is correctly captured as `3` because `cur` is updated before `best` is compared on every iteration, including the final one.

out 845 tok (+596 thinking) latency 3.5 s

Whether the fix is actually correct (runnable), explanation density, and token efficiency on a bounded task.

PROMPT Extract a JSON object with fields {company, quarter, revenue_usd_m, growth_pct, guidance} from this text. Use null for anything not stated; add no other fields. CHECK valid JSON, schema exact

Kimi K2.7 Code passed · 5/5 fields, guidance "withheld pending the Q4 close"

```json { "company": "Meridian Robotics", "quarter": "September quarter", "revenue_usd_m": 412, "growth_pct": 11, "guidance": "withheld pending the Q4 close" } ```

out 721 tok (+663 thinking) latency 13.2 s

MiniMax M3 passed · 5/5 fields, guidance null

```json { "company": null, "quarter": "Q3", "revenue_usd_m": 412, "growth_pct": null, "guidance": null } ```

out 1002 tok (+955 thinking) latency 3.6 s

Schema obedience (no invented fields), hallucination pressure (guidance is explicitly withheld), and structured-output path differences.

PROMPT Write a 120-word product announcement for an API gateway feature that caches prompts across providers. Forbidden words: "seamless", "unlock", "game-changer", "revolutionize", "empower". Exactly one sentence must be a question. CHECK 120 words, 0 banned words

Kimi K2.7 Code passed · 120 words, 0 banned, 1 question

We're introducing Prompt Cache for our API Gateway, a new capability that stores prompt responses across multiple AI providers. By caching replies, you reduce redundant calls, lower latency, and cut costs without changing how you build. Teams using several model providers can now reuse identical prompt results instantly, keeping traffic consistent and budgets predictable. Responses are keyed by provider, model, and exact prompt, so you always get the right result. Configure TTL, hit thresholds, and eviction rules from a single dashboard. It fits into your existing routing and requires no code changes. Setup takes minutes and works with your current endpoints. Want to see how much latency and spend you can trim? Check the docs to enable Prompt Cache today.

out 2375 tok (+2235 thinking) latency 38.6 s

MiniMax M3 passed · 121 words, 0 banned, 1 question

Introducing PromptCache, our intelligent caching layer for the API gateway. Tired of paying twice for identical requests? The feature stores prompt-response pairs across every connected provider, returning cached results instead of hitting costly endpoints. Deployment requires only a single configuration flag, and semantic similarity detection happens automatically. Latency drops up to 70% on repeated traffic while costs follow the same downward curve. Developers access detailed cache hit metrics, eviction logs, and provider breakdowns through the dashboard. Whether routing to OpenAI, Anthropic, or local models, every cached token saves time and money. The cache respects TTL policies, supports manual purging, and encrypts stored content. Available today for Pro and Enterprise tiers—enable PromptCache on existing endpoints and watch inference bills shrink immediately.

out 2808 tok (+2656 thinking) latency 6.0 s

Constraint obedience (word budget, banned-word list, the single question), style fingerprint, and length control.

Switch between them with one line

Both ids are in every tab below; the highlighted pair of lines is the only edit. Same endpoint, same key, same request shape.

from openai import OpenAI

client = OpenAI(
    base_url="https://synthorai.io/v1",
    api_key="sk-syn-...",
)

resp = client.chat.completions.create(
    model="kimi-k2.7-code",
    # model="minimax-m3",  # uncomment this line, comment the one above
    messages=[{"role": "user", "content": "Summarize this diff"}],
    reasoning_effort="medium",
)
print(resp.choices[0].message.content)

Get your API key →

FAQ

Which is cheaper, Kimi K2.7 Code or MiniMax M3?

MiniMax M3 is cheaper on the "Input / 1M tokens" row ($0.3 vs $0.95, 3.2× apart). Other rows may point the other way; the table above carries the full rate card, and real cost depends on your mix.

Can I A/B test Kimi K2.7 Code against MiniMax M3 without two integrations?

Yes. Both are served through the same OpenAI-compatible endpoint with one API key. Switching is a one-line change to the model id, so you can route a fraction of traffic to each and compare bills directly.

Do Kimi K2.7 Code and MiniMax M3 support prompt caching?

Yes. Both bill cache reads below their input rate, so warm-prefix workloads cost less than the list rates suggest. The exact cache-read rows are in the pricing table above.

Related comparisons

From our measured studies