MiniMax-M2 runs coding tasks cheaply: a real 38-in/200-out coding call billed $0.0002514 in 2.95s, measured via OpenRouter. DeepSeek V4-Flash is roughly half the per-token rate, so the right pick depends on whether you value M2's output quality or DeepSeek's raw cost floor. That tradeoff, not the headline price, is what decides a coding bill.
This piece is part of the MiniMax API pricing hub, and it focuses on one question buyers actually ask: what does it cost to run MiniMax-M2 on coding work, and how does that compare to the cheapest credible alternative, DeepSeek V4-Flash.
MiniMax-M2 is a general-purpose Chinese LLM that handles coding tasks at a price point well below US frontier models. Here is what it actually costs per million tokens, and what one real coding call billed.
According to MiniMax, MiniMax-M2 official pricing is $0.30 per million input tokens and $1.20 per million output. The OpenRouter-routed rate we benchmark against runs slightly cheaper at $0.255 input and $1.00 output. We mention both because the number you pay depends entirely on how you reach the model, and the official figure still needs native-key re-verification.
For comparison, according to DeepSeek API Docs, DeepSeek V4-Flash charges $0.14 input and $0.28 output per million tokens. That is the cheapest credible coding rate in this comparison, and it is the reason DeepSeek keeps showing up on every cost-conscious shortlist.
| Model | Input ($/1M) | Output ($/1M) | Routing | Source |
|---|---|---|---|---|
| MiniMax-M2 | $0.255 | $1.00 | OpenRouter-routed | OpenRouter catalog |
| MiniMax-M2 (official) | $0.30 | $1.20 | platform.minimax.io | MiniMax pricing (reverify) |
| DeepSeek V4-Flash | $0.14 | $0.28 | official api.deepseek.com | DeepSeek API Docs |
The gap that matters for coding is the output rate. Coding workloads are output-heavy, since a model that writes a function or a patch emits far more tokens than the few it reads in. At $1.00 versus $0.28 per million output, M2 is roughly 3.5 times the output cost of V4-Flash. That single ratio drives most of the per-task spread you see below.
Documentation gives you the rate card. It does not tell you what a coding turn actually bills or how fast it returns, so we called the model and measured it.
When we sent a coding prompt to MiniMax-M2 through OpenRouter, the call consumed 38 input tokens and produced 200 output tokens, billed at $0.0002514, and it returned in 2.95 seconds. That is the measured cost of one short coding turn, routed via OpenRouter and live-billed, not an estimate from the rate card.
We ran a DeepSeek V4-Flash coding call as the anchor. According to DeepSeek API Docs, V4-Flash bills that call at $0.14 input and $0.28 output per million tokens. Hitting the official api.deepseek.com endpoint directly, the call used 27 input and 86 output tokens and returned in 1.96 seconds. Two things stand out. DeepSeek was faster, 1.96s against 2.95s, and it emitted fewer output tokens for its answer, which compounds its already-lower output rate into a real per-task saving. One caveat we disclose plainly: the MiniMax number is OpenRouter-routed while the DeepSeek number is from the native endpoint, so this is a routed-versus-official comparison, not two calls on identical infrastructure.
| Model | Tokens (in/out) | Latency | Routing |
|---|---|---|---|
| MiniMax-M2 | 38 / 200 | 2.95s | OpenRouter (live-billed) |
| DeepSeek V4-Flash | 27 / 86 | 1.96s | official api.deepseek.com |
A single 200-token reply costs a fraction of a cent on either model, so the headline number feels free. It stops feeling free at volume. The useful unit is cost per thousand coding tasks, where each task reads a chunk of context and writes a meaningful patch.
Take a representative coding task: 2,000 input tokens of context plus 800 output tokens of generated code. On MiniMax-M2 at the OpenRouter rate that is 2,000 x $0.255/1M plus 800 x $1.00/1M, or $0.00051 + $0.00080 = $0.00131 per task. On DeepSeek V4-Flash at $0.14/$0.28 the same task is $0.00028 + $0.000224 = $0.000504. Across 10,000 tasks that is roughly $13.10 on M2 against $5.04 on V4-Flash.
| Per coding task (2K in / 800 out) | MiniMax-M2 (OpenRouter) | DeepSeek V4-Flash (official) |
|---|---|---|
| Input cost | $0.00051 | $0.00028 |
| Output cost | $0.00080 | $0.000224 |
| Per-task total | $0.00131 | $0.000504 |
| Per 10,000 tasks | ~$13.10 | ~$5.04 |
The practical upshot is that DeepSeek V4-Flash runs a coding workload at roughly 38% of MiniMax-M2's cost on identical token assumptions. If your only metric is dollars per token, DeepSeek wins this row outright, and it is not close.
Cost is one axis, not the whole decision. The reason a buyer would pay M2's premium is output quality and behavior on harder coding tasks, where a model that one-shots a correct patch beats a cheaper model that needs three tries. Three failed cheap attempts cost more than one good expensive one, in both tokens and engineer time.
Choose DeepSeek V4-Flash when your coding tasks are well-scoped and high-volume, where the 38%-of-cost figure dominates and per-task quality differences are small. Choose MiniMax-M2 when task difficulty is the bottleneck and a higher first-pass success rate saves more than the extra output spend. The 2.95s latency we measured is fine for batch and agent loops; if you need sub-2s interactive turns, DeepSeek's 1.96s in our test was the quicker of the two.
According to MiniMax, M2's pricing sits below US frontier models like GPT-4o and Claude Sonnet, so even at the higher end of this comparison it remains a low-cost option in absolute terms. The contest here is not MiniMax versus an expensive incumbent; it is two cheap Chinese models splitting on a cost-versus-quality line.
How much does a MiniMax-M2 coding call cost? In our live test, one coding call of 38 input and 200 output tokens billed $0.0002514 and returned in 2.95 seconds, measured via OpenRouter. At the per-million rate that is $0.255 input and $1.00 output on OpenRouter, or $0.30 / $1.20 at MiniMax's official rate, which still needs native-key re-verification.
Is MiniMax-M2 cheaper than DeepSeek for coding? No. On identical token assumptions DeepSeek V4-Flash at $0.14 / $0.28 runs a coding workload at roughly 38% of MiniMax-M2's cost. The gap comes mostly from the output rate, $0.28 versus $1.00 per million, and coding is output-heavy.
Which is faster for coding, MiniMax-M2 or DeepSeek V4-Flash? In our tests DeepSeek V4-Flash returned in 1.96 seconds on the official endpoint versus 2.95 seconds for MiniMax-M2 via OpenRouter. Both are fine for batch and agent loops; DeepSeek had the edge on interactive latency.
Why pay more for MiniMax-M2 if DeepSeek is cheaper? Because per-task cost is not just per-token price. If M2 produces a correct patch on the first pass where a cheaper model needs several retries, the higher output rate can still come out ahead in total tokens and engineer time. Match the model to task difficulty, not the rate card alone.
This is part of the MiniMax API pricing hub. For DeepSeek's full coding rate card, see the DeepSeek API pricing hub.
Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing. Last verified: 2026-06-26.
<!-- METADATA { "title": "MiniMax-M2 Coding API Cost in Practice (2026)", "slug": "minimax-coding-cost", "meta_description": "We tested it live via OpenRouter: a MiniMax-M2 coding call billed $0.0002514 in 2.95s. Per-task cost table vs DeepSeek V4-Flash, plus the cost-quality tradeoff.", "focus_keyword": "minimax coding cost", "secondary_keywords": ["minimax-m2 coding api cost", "minimax vs deepseek coding cost", "minimax m2 api price", "cheapest coding llm api"], "tags": ["MiniMax", "API Pricing", "Coding", "DeepSeek"], "category": "Pricing", "cluster_id": "minimax-api-pricing", "cluster_role": "micro", "hub_slug": "minimax-api-pricing", "evidence_file": "clients/china-llm-aggregator/articles/minimax-api-pricing-evidence.json", "needs_native_reverify": true, "verified_until": "2026-09-24", "author_name": "Kevin Fan", "author_title": "Customer Success Manager", "author_linkedin": "", "author_expertise": ["Chinese LLM ecosystem", "AI infrastructure pricing", "model benchmarking", "cross-border AI compliance"], "faq_pairs": [ {"q": "How much does a MiniMax-M2 coding call cost?", "a": "In our live test, one coding call of 38 input and 200 output tokens billed $0.0002514 and returned in 2.95 seconds, measured via OpenRouter. At the per-million rate that is $0.255 input and $1.00 output on OpenRouter, or $0.30 / $1.20 at MiniMax's official rate, which still needs native-key re-verification."}, {"q": "Is MiniMax-M2 cheaper than DeepSeek for coding?", "a": "No. On identical token assumptions DeepSeek V4-Flash at $0.14 / $0.28 runs a coding workload at roughly 38% of MiniMax-M2's cost. The gap comes mostly from the output rate, $0.28 versus $1.00 per million, and coding is output-heavy."}, {"q": "Which is faster for coding, MiniMax-M2 or DeepSeek V4-Flash?", "a": "In our tests DeepSeek V4-Flash returned in 1.96 seconds on the official endpoint versus 2.95 seconds for MiniMax-M2 via OpenRouter. Both are fine for batch and agent loops; DeepSeek had the edge on interactive latency."}, {"q": "Why pay more for MiniMax-M2 if DeepSeek is cheaper?", "a": "Because per-task cost is not just per-token price. If M2 produces a correct patch on the first pass where a cheaper model needs several retries, the higher output rate can still come out ahead in total tokens and engineer time. Match the model to task difficulty, not the rate card alone."} ], "external_links_used": [ {"url": "https://platform.minimax.io", "source_name": "MiniMax", "claim": "MiniMax-M2 official pricing $0.30/$1.20 per 1M tokens; sits below US frontier models (needs native reverify)"}, {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs", "claim": "DeepSeek V4-Flash $0.14 input / $0.28 output per 1M tokens"} ], "internal_links_used": [ {"url": "/blog/minimax-api-pricing/", "anchor_text": "MiniMax API pricing hub", "type": "hub"}, {"url": "/blog/deepseek-api-pricing/", "anchor_text": "DeepSeek API pricing hub", "type": "cross-cluster"} ], "first_hand_evidence": { "source": "minimax-api-pricing-evidence.json runs minimax-m2_coding + deepseek_v4flash_coding_anchor", "measured": "MiniMax-M2 coding (OpenRouter): 38 in / 200 out, $0.0002514, 2.95s. DeepSeek V4-Flash coding anchor (official): 27 in / 86 out, 1.96s.", "captured": "2026-06-26", "disclosure": "MiniMax measured via OpenRouter; DeepSeek via official api.deepseek.com" }, "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial comparison diagram of MiniMax-M2 vs DeepSeek V4-Flash per-coding-task cost, annotated with the live test result of a 38-in/200-out MiniMax-M2 call billing $0.0002514 in 2.95s via OpenRouter. Blue and teal palette. Minimal background. 16:9.", "alt": "Comparison diagram of MiniMax-M2 versus DeepSeek V4-Flash per-coding-task API cost, annotated with a live MiniMax-M2 coding call billing $0.0002514 in 2.95 seconds via OpenRouter"} ] } -->