A Shanghai AI company founded in 2021 and listed on the Hong Kong Stock Exchange in January 2026. MiniMax builds foundation models, API products, and consumer apps.
The June 2026 portfolio is now led by MiniMax-M3 for text/coding/agentic work, with MiniMax Code as the paired coding-agent surface. The same company also operates Hailuo 2.3 video generation, Speech 2.8 for current voice APIs, Music 2.6 for music, and Talkie for companion-character chat.
System Verdict
Pick MiniMax if you want to benchmark M3 as a low-cost, long-context, multimodal coding/agent model. supports up to a 1M-token context window with a guaranteed minimum of 512K tokens.
Skip it if procurement needs ecosystem maturity, independent benchmark proof, or Western data-residency defaults. MiniMax publishes aggressive M3 benchmark claims, but production buyers should test it against Claude, ChatGPT, Gemini, Qwen, Kimi, and GLM on their own tasks before moving workloads.
Do not buy from an old M2.7 mental model. M2.7 still appears in the pricing table and older docs, but M3 is the current flagship path for new model evaluation as of June 21, 2026.
Key Facts
| Founded | 2021, Shanghai |
| Public listing | Hong Kong Stock Exchange, January 2026 |
| Current flagship text model | MiniMax-M3 |
| M3 context | Up to 1M tokens; guaranteed minimum 512K tokens in the official M3 API positioning |
| M3 standard price | <=512K input: $0.30/M input, $1.20/M output, $0.06/M prompt-cache read |
| M3 >512K input | $0.60/M input, $2.40/M output, $0.12/M prompt-cache read; listed with limited-access caveats |
| M3 Priority tier | <=512K input: $0.45/M input, $1.80/M output; >512K input: $0.90/M input, $3.60/M output |
| Older text models still listed | M2.7, M2.7-highspeed, M2.5, M2.5-highspeed, M2.1, M2.1-highspeed, M2 |
| Speech | Speech 2.8 HD/Turbo current in API docs; Speech 2.6, Speech-02, and Speech-01 remain supported in T2A HTTP docs |
| Video | Hailuo 2.3 / Hailuo 2.3 Fast |
| Music | Music 2.6 |
| Consumer apps | MiniMax Agent / MiniMax Code and Talkie |
What it actually is
Developer API. work, plus older M2 models still visible in pricing/docs. The platform also exposes Speech, Hailuo video, image, music, and MCP-vlm pricing under separate billing routes.
MiniMax Code and MiniMax Agent. The current product push is agentic coding and long-context workflows. MiniMax positions M3 as the model trained to pair with MiniMax Code.
Hailuo, Speech, and Music. Adjacent generation APIs under the same company. Hailuo is covered on Hailuo; voice is covered on MiniMax Speech.
Talkie. A character-chat and companion app. It proves consumer appetite, but it also carries moderation and copyright risk around public-figure/persona simulations.
When to pick MiniMax
- M3 API evaluation. Benchmark M3 when the brief is low-cost coding, agentic, long-context, or multimodal input work.
- Cost-sensitive model diversification. MiniMax belongs beside Qwen, Kimi, GLM, and Mistral in non-OpenAI model evaluations.
- Multimodal vendor consolidation. Text, voice, video, music, and MCP/API-vlm pricing sit under one developer platform, though the billing lanes differ.
- Voice app builders. Speech 2.8 HD/Turbo, voice cloning, streaming T2A, and long-form async speech generation are strong reasons to shortlist MiniMax.
- Companion-chat products. Talkie gives MiniMax direct consumer feedback loops for character and persona workflows.
When to pick something else
- Most mature English assistant: ChatGPT or Claude.
- Google-stack integration: Gemini for Workspace, Google AI subscriptions, and Google Cloud adjacency.
- Open China-model ecosystem: Qwen when Alibaba Cloud, Apache-licensed Qwen3 weights, and Qwen Chat are the strategic fit.
- US/EU data-residency default: OpenAI, Anthropic, Google, or Mistral are cleaner starting points for many regulated Western teams.
- Premium audiobook or studio voice: ElevenLabs. MiniMax Speech wins on API economics; ElevenLabs still wins on creator workflow maturity and quality ceiling.
Pricing
MiniMax-M3 pay-as-you-go text API (per 1M tokens):
| M3 lane | Input | Output | Prompt-cache read | Notes |
|---|---|---|---|---|
| Standard, <=512K input | $0.30 | $1.20 | $0.06 | Main June 2026 buyer anchor |
| Standard, >512K input | $0.60 | $2.40 | $0.12 | Docs say limited quantity / limited-time access |
| Priority, <=512K input | $0.45 | $1.80 | $0.09 | Enabled through service_tier; access caveats apply |
| Priority, >512K input | $0.90 | $3.60 | $0.18 | Access caveats apply |
Older text-model pricing still listed: M2.7 remains at $0.30/M input and $1.20/M output, while M2.7-highspeed remains at $0.60/M input and $2.40/M output. Treat these as compatibility/fallback lanes unless your workload specifically needs M2 behavior.
Voice, video, music APIs: separate pricing. June 21 pay-as-you-go docs list Speech 2.8 Turbo at $60/M characters, Speech 2.8 HD at $100/M characters, Hailuo 2.3 at $0.28 for a 768P 6-second clip, Hailuo 2.3 Fast at $0.19 for a 768P 6-second clip, Music 2.6 at $0.15 per up-to-5-minute generation, image-01 at $0.0035/image, and API-vlm at $0.06/request.
Prices verified 2026-06-21 via the MiniMax pay-as-you-go pricing docs. Token Plan, Credits, Audio Subscription, Video Packages, and pay-as-you-go are different purchase paths; do not assume credits or quotas move between them.
Against the alternatives
| MiniMax M3 | Claude / ChatGPT / Gemini | Qwen | Kimi / GLM | |
|---|---|---|---|---|
| Best reason to test | Low-cost M3 coding, agentic, multimodal, long-context API | Mature Western frontier assistants and ecosystems | Alibaba/Qwen Cloud and open-weight Qwen family | China/Asia model diversification and long-context APIs |
| Buyer proof needed | Independent task benchmarks, availability, data residency | Plan fit, enterprise controls, price/performance | Cloud fit, model license, API terms | Current model/version path and pricing |
| Context positioning | Up to 1M, guaranteed minimum 512K in M3 API positioning | Varies by provider/model | Long-context model family | Long-context model families |
| Voice/video adjacency | Yes: Speech 2.8 and Hailuo | Usually separate products/providers | Mixed ecosystem | Mixed ecosystem |
| Main risk | Docs and access tiers are moving quickly | Higher cost / product constraints | Procurement and ecosystem fit | Fast-moving versions and regional posture |
Failure modes
- Vendor benchmark claims need replication. MiniMax’s official M3 pages make strong coding, browsing, multimodal, and agentic benchmark claims. Treat them as vendor claims until your own evaluation confirms them.
- 512K versus 1M access matters. The official model page says M3 supports up to 1M context with a guaranteed minimum of 512K. The pricing page separately flags >512K input as limited/early access. For production planning, assume 512K until your account proves otherwise.
- Priority tiers are not ordinary self-serve yet. The pricing page lists
service_tierPriority pricing, but also says access is gated/early. Do not build an SLA around Priority until procurement confirms it. - Billing surfaces are easy to mix up. Token Plan, Credits, Audio Subscription, Video Packages, and pay-as-you-go are separate routes.
- M2 docs still exist. The API overview and older text-generation docs still expose M2.7/M2.5 lanes. New buyers should start with M3, but integrations may find stale examples.
- Data residency is China-first. Enterprise compliance in regulated US and EU sectors requires careful review or a different vendor.
- Talkie carries moderation and copyright risk. Companion-character products are commercially useful but legally sensitive, especially around public figures and entertainment IP.
- English-language community support is thinner. API docs exist in English, but troubleshooting resources and third-party examples are not as deep as OpenAI, Anthropic, Google, or Mistral.
Methodology
This page was rechecked by the aipedia.wiki editorial workflow on June 21, 2026 against the MiniMax M3 model page, MiniMax M3 launch post, M3 for AI Coding Tools docs, MiniMax pay-as-you-go pricing, MiniMax pricing overview, MiniMax T2A API overview, and MiniMax company/financial-results sources. Scoring follows the four-dimension rubric at /about/scoring/ (Utility x Value x Moat x Longevity, unweighted average).
FAQ
Is MiniMax free to use? Consumer MiniMax products can be tried through public product surfaces, and the developer platform supports several purchase routes. API procurement should start by choosing between Token Plan/Credits and pay-as-you-go; the June 21 pay-as-you-go table lists MiniMax-M3 standard at $0.30/M input and $1.20/M output for <=512K input tokens.
What is the current MiniMax flagship model? MiniMax-M3. It was released June 1, 2026 and is now the current flagship model path for coding, agentic, long-context, and native multimodal evaluation. M2.7 remains visible in pricing/docs but should no longer be treated as the primary new-buyer benchmark.
Does MiniMax-M3 really support 1M context? MiniMax’s official M3 page says the API supports up to 1M tokens with a guaranteed minimum of 512K. The pay-as-you-go pricing page flags >512K input as limited/early access. For buyer math, verify your account’s actual context tier before designing around 1M.
How does MiniMax relate to Hailuo AI? Hailuo is MiniMax’s text-to-video usage. See the Hailuo page for video-specific buyer guidance.
Is MiniMax available outside China? Yes, MiniMax exposes international product and API surfaces. Compliance teams should still review data residency, contract terms, and private-deployment options before regulated workloads.
What is Talkie? Talkie is MiniMax’s character and companion-chat app. It is strategically important because it gives MiniMax consumer-scale persona-chat data and product feedback, but it also carries moderation, safety, and copyright risk.
Sources
- MiniMax M3 model page: M3 positioning, context, multimodality, API access, and MiniMax Code path (verified 2026-06-21)
- MiniMax M3 launch post: June 1 release details and vendor benchmark claims (verified 2026-06-21)
- M3 for AI Coding Tools: coding-tool setup and model naming (verified 2026-06-21)
- MiniMax pay-as-you-go pricing: current text, audio, video, music, image, and MCP usage rates (verified 2026-06-21)
- MiniMax platform pricing: Token Plan, Credits, Audio Subscription, Video Packages, and pay-as-you-go route overview
- MiniMax T2A API overview: current Speech 2.8 and voice API surface
- MiniMax FY2025 results: company and multimodal usage context
Related
- Category: AI Chatbots · AI Research
- Siblings: Hailuo · MiniMax Speech
- Compare: Claude · ChatGPT · Gemini · Qwen
