Space Bunny Revealed: MiniMax M3.1 Flash Specs, Pricing, and API Guide

What is Space Bunny Alpha? MiniMax confirms the identity

Space Bunny Alpha appeared on OpenRouter and OpenCode in September 2026 under anonymous labels such as stealth/space-bunny-alpha and space-bunny-free. Developers used it for repository edits, coding agents, UI work, tool calls, and long multi-step sessions.

MiniMax's October 19 launch post explicitly states: “Space Bunny 🐰, which processed more than 200 trillion tokens while operating anonymously, was an early version of MiniMax M3.1 Flash.” Space Bunny is therefore a historical preview identity for an early version, rather than a separate vendor or an exact guarantee that every preview checkpoint equals the final release.

Space Bunny specs: MiniMax M3.1 Flash model card

The output limit is a serving limit, not a promise that a request can contain 1M input tokens plus 524,288 output tokens. Budget input, tool messages, reasoning, and output against the selected endpoint's context and token limits.

Why people thought Space Bunny was MiniMax

Before the official reveal, community tokenizer probes reportedly matched MiniMax at 24/24 and 50/50, while one GLM comparison matched only 7/24. A large context window, multimodal input, five effort controls, and the September MiniMax Code preview also pointed toward MiniMax.

Those observations were useful historical clues, but tokenizer compatibility could not identify an exact checkpoint. They no longer need to carry the identity claim: the October 19 vendor announcement supplies direct confirmation. The GLM theory should not be repeated as the current explanation.

Space Bunny versus MiniMax M3.1 Flash: what changed?

There is also a distinct model called MiniMax M3. Its APIMaster ID minimax-m3 does not invoke M3.1 Flash. A gateway key or similar model name does not make the two interchangeable.

Space Bunny benchmark update: M3.1 Flash results

MiniMax reports the following launch results. These are vendor-reported scores, not APIMaster's independently reproduced tests:

MiniMax describes OSWorld V2 and ALE as leading among Chinese models at launch. OpenRouter's current catalog separately links an Artificial Analysis Intelligence Index score of 42.7. The suites, harnesses, and scoring units differ; these numbers should not be combined into a single success rate.

The vendor also claims nearly 3× faster inference, everyday peaks above 200 TPS, median P50 above 170 TPS, and 70% lower overall agentic usage cost than its predecessor. The 70% figure concerns agent workloads, not a universal API-token discount. Gateway queues, tool latency, effort level, and reasoning length can produce different end-to-end results.

Community feedback: what the preview showed

September users valued free access, speed, and coherent multi-file work. Historical examples included small applications, drag-and-drop bug fixes, screenshot-driven tools, and SVG or UI generation. Preview reports of roughly 70–120 TPS describe those routes and dates; they are not the official release's guaranteed speed.

The anonymous preview's traffic and the vendor's reported 200T-token total show substantial usage. Neither is a substitute for testing whether the released endpoint can complete your own repository tasks reliably.

Complaints and limits: what still needs testing

Preview complaints included overthinking on simple prompts, strict-format failures, uneven small coding tasks, and interruptions on free routes. One historical independent suite reportedly returned 128/157; another anecdote described 129 lines instead of six. These are dated preview observations, not fresh M3.1 Flash benchmark results.

For the current release, measure valid tool calls, patch acceptance, retry rate, wall-clock time, and cost per completed task. Check provider data-handling terms before sending private repositories. A no-training statement on an old OpenRouter preview does not establish the retention or training policy of every current endpoint.

Space Bunny effort settings and agent workflow

M3.1 Flash always reasons. In MiniMax's OpenAI-compatible API, use reasoning_effort; omitting it defaults to max. reasoning_effort: "none" or thinking: {"type":"disabled"} returns an error. Other gateways can expose different parameter shapes.

Preserve the reasoning and tool-call fields required by your provider when carrying conversation history forward. Set timeouts and output limits, validate structured output, and keep side-effecting tools idempotent. These controls matter more than simply selecting the highest effort for every request.

Space Bunny API and OpenCode migration

If OpenCode reports “Model space-bunny-free has been deprecated. Use minimax-m3.1-flash-free instead.”, update the model selected for the OpenCode Zen provider. Check agent-specific overrides as well as your main model configuration.

These identifiers belong to different providers. Do not paste the OpenCode ID into MiniMax's API or assume an APIMaster key authorizes an OpenRouter request.

For the released Flash model, the official OpenAI-compatible request is:

curl "https://api.minimax.io/v1/chat/completions" \
  -H "Authorization: Bearer $MINIMAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"MiniMax-M3.1-Flash","reasoning_effort":"medium","messages":[{"role":"user","content":"Inspect this failing test and propose a minimal patch."}]}'

For the currently available M3 alternative on APIMaster, check the M3 price and route card, then use:

curl "https://apimaster.ai/v1/chat/completions" \
  -H "Authorization: Bearer $APIMASTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"minimax-m3","messages":[{"role":"user","content":"Inspect this failing test and propose a minimal patch."}]}'

Free access, weights, and what to verify next

At the October 20 check, OpenRouter's catalog lists the free Flash route with zero prompt and completion prices. It does not expose an exact free-offer expiry. Treat claims of a 72-hour launch offer as something to verify against current campaign terms; do not promise an October 22 cutoff without a confirmed start time and time zone. Free routes can have separate quotas and capacity limits.

MiniMax also lists official API access, M Plan, and MiniMax Code desktop access. Subscription entitlements and API token billing are separate products. The launch post says weights will be released later; it does not confirm 428B parameters or an October 23 release date.

APIMaster now lists two active public MiniMax-M3.1-Flash routes, checked October 20, 2026. Both cost $0.15 input / $0.60 output per million tokens, matching MiniMax’s standard price (0% discount). Check the live model card for limits; the marketplace card is authoritative.

FAQ

What is Space Bunny Alpha?

Space Bunny Alpha was an anonymous preview identity for an early MiniMax M3.1 Flash version. MiniMax confirmed that relationship in its October 19, 2026 launch announcement.

Is Space Bunny MiniMax M3.1 Flash or GLM?

MiniMax confirms it was an early M3.1 Flash version. The former GLM attribution theory is superseded by the official announcement; preview and release checkpoints need not be identical.

What are the Space Bunny context and output limits?

M3.1 Flash has a documented 1,000,000-token context window. OpenRouter currently advertises up to 524,288 output tokens. Input, reasoning, and output must still fit the endpoint's token budget.

Is Space Bunny free, and when does the offer end?

The current OpenRouter model is minimax/minimax-m3.1-flash:free, listed at zero prompt and completion prices on October 20. Its live catalog does not publish an exact expiry; check campaign terms and account quotas rather than assuming a fixed 72-hour window.

How do I fix the OpenCode space-bunny-free deprecation error?

For OpenCode Zen, replace the old space-bunny-free selection with minimax-m3.1-flash-free, including agent overrides. Other providers require their own IDs and credentials.

What does MiniMax M3.1 Flash cost?

MiniMax's standard API price is $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens. That is official pricing, not an APIMaster quote or a permanent OpenRouter free-access guarantee.

Can I use M3.1 Flash on APIMaster?

APIMaster now lists two active public MiniMax-M3.1-Flash routes, checked October 20, 2026. Both cost $0.15 input / $0.60 output per million tokens, matching MiniMax’s standard price (0% discount). Check the live model card for limits; the marketplace card is authoritative.

Can I disable M3.1 Flash reasoning?

No. The official API always enables thinking and rejects disabled thinking or reasoning_effort: "none". Select low, medium, high, xhigh, or max; the default is max.

Sources and current status

• MiniMax M3.1 Flash launch announcement — October 19 identity confirmation, speed, costs, benchmarks, and weights planned later.
• MiniMax model and text-generation guide — 1M context, official model ID, and reasoning controls.
• MiniMax OpenAI-compatible API guide — request fields, always-on thinking, and multimodal support.
• MiniMax official pricing — Flash pricing and M3 context-dependent tiers.
• OpenRouter M3.1 Flash free listing and model catalog — current free ID, modality limits, output cap, and AA-linked score.
• OpenCode Zen model catalog — current minimax-m3.1-flash-free ID.
• APIMaster live M3 card — three active routes and price range checked October 20, 2026; live card takes precedence.

To test the available M3 routes, register with APIMaster, create an API key, set your OpenAI-compatible base URL to https://apimaster.ai/v1, and select minimax-m3. Compare route limits before using the up to ~50% discount against the official >512K M3 reference; the minimum equals the short-context standard price.