MiniMax

MiniMax M3.1 Flash

MiniMax M3.1 Flash is $0.15/$0.60 per million tokens on MiniMax's API — half of M3's ≤512K rate — with a 1M context and no length step (checked 2026-10-19). Artificial Analysis HLE without tools is 47.96 against M3's AA 39.0, an 8.96-point real gap past the 2-point band; Terminal-Bench 4.0 is 25.76. OpenCode's 2026-10-19 commit names Space Bunny Free as this model.

MiniMax M3.1 Flash benchmarks and pricing, every number sourced: 2 tracked MiniMax M3.1 Flash benchmark scores (2 independently run, 0 still resting on a vendor’s own claim), priced at $0.15 per million input tokens and $0.60 per million output.

Released: 2026-10-19
License: proprietary
Context window: 1M tokens
Knowledge cutoff: Not disclosed
Verified: source checked 2026-10-19
Parameters: Not disclosed
Architecture: Not disclosed

MiniMax M3.1 Flash’s verified record

MiniMax M3.1 Flash’s most-compared rival is GLM-5.3-Flash: 1 lead and 1 tie across their 2 shared comparisons. MiniMax M3.1 Flash is priced at $0.15/$0.60 per 1M tokens (in/out) vs GLM-5.3-Flash’s $0.15/$0.50.

Against the 45 head-to-head comparisons MiniMax M3.1 Flash shares with other tracked models:

27 real gaps11 inside the band7 not called

A gap counts for MiniMax M3.1 Flash only where both sides were run independently and the benchmark still separates models. Losses are listed alongside wins on purpose — MiniMax M3.1 Flash trails on 9 of them.

HLE · no tools Reasoning

±2 is noise

Behind

Muse Wonder −2.1 · GPT-6.1 Sol −4.9 · GPT-6 Astra −6.7 · Claude Sonnet 5.5 −7.0 · Gemini 4 Argon −9.1 · Claude Fable 5.1 −11.1 · Claude Opus 5.5 −13.4 · 2 previous: Claude Opus 5 −6.9 · Claude Fable 5 −7.5

Ahead

Claude Haiku 5.5 +3.6 · Muse Spark Open +4.1 · DeepSeek V4.1 Flash +4.6 · Qwen3.8-Max +4.9 · Grok 4.7 +4.9 · GLM-5.3 +5.7 · DeepSeek V4 Pro (0813) +7.0 · GLM-5.3-Flash +8.1 · MiniMax M3 +9.0 · GPT-6 Luna +9.5 · Qwen3.8-Flash-Next +10.0 · Mistral Large 4 +12.9 · 6 previous: Muse Spark 1.2 +2.5 · Grok 4.6 +5.1 · Claude Sonnet 5 +6.7 · GLM-5.2 +6.9 · GPT-5.6 Luna +8.5 · DeepSeek V4 Flash (0731) +9.4

Tie

10 models within ±2

Unverified

1 model — vendor-reported on one side

No verdict for MiniMax M3.1 Flash anywhere on Terminal-Bench 4.0 ( every independently confirmed comparison inside the noise band).

No real gap yet in any of MiniMax M3.1 Flash’s agentic comparisons — the independently confirmed ones all sit inside the noise band.

MiniMax M3.1 Flash API pricing

$0.15 in / $0.60 out per 1M tokens — official pricing source

What MiniMax M3.1 Flash costs per job

MiniMax M3.1 Flash cost for three reference workloads, computed from its list rates
Workload	Tokens in / out	Cost
One long chat turn	100K / 10K	$0.021
A codebase review	1,000K / 100K	$0.210
A day of agent work	10,000K / 1,000K	$2.10

Computed from MiniMax M3.1 Flash’s list rates above — cache discounts and batch tiers are not applied.

MiniMax M3.1 Flash is one of 2 MiniMax models tracked on this site, at these official list prices.

MiniMax model pricing, official list rates
Model	In / 1M	Out / 1M
MiniMax M3.1 Flash	$0.15	$0.60
MiniMax M3	$0.30	$1.20

MiniMax M3.1 Flash benchmark scores

MiniMax M3.1 Flash benchmark scores, provenance, and source links
Benchmark	Score
HLE(no tools)^[1] Reasoning · ±2 is noise	48.0artificialanalysis.ai · 2026-10-19
Terminal-Bench 4.0^[2] Terminal ops · ±12.4 is noise	25.8artificialanalysis.ai · 2026-10-19

Who ran these numbers: 2 of 2 independent — artificialanalysis.ai (2).

HLE: AA embedded hle 0.479610750695088 → 47.96. Display name has no effort tier.
Terminal-Bench 4.0: AA embedded terminalBench40 0.257575757575758 → 25.76.

Notes on the record

MiniMax's platform pricing page (checked 2026-10-19) lists MiniMax-M3.1-Flash at $0.15 input / $0.60 output per million tokens, with prompt-cache reads at $0.015, and no >512K surcharge of the kind MiniMax M3 still carries. Context is marketed at 1M tokens on the models intro and text-generation guide. No Hugging Face repo for MiniMax-M3.1-Flash was returned by the HF models API on 2026-10-19, and the open-weight phrasing on the models page attaches to MiniMax-M3, so this record stays proprietary until weights are published under that name.

Artificial Analysis's minimax-m3-1-flash page (checked 2026-10-19) embeds HLE 0.479611→47.96 and terminalBench40 0.257576→25.76 with no effort tier in the display name. MiniMax M3's AA HLE without tools on this site is 39.0; the 8.96-point gap exceeds HLE's 2.0 meaningful_gap on a matched independent runner. M3's stored Terminal-Bench 4.0 cell is vals.ai's 1.01, a different harness, so those two cells are not a clean pair. On Artificial Analysis itself, M3 reads 2.02 with no effort tag (checked 2026-10-19); 25.76 minus 2.02 is 23.74, past the 12.4 band.

OpenCode's models.dev commit fbfb5229 (2026-10-19T07:37:37Z), message "Reveal Space Bunny as MiniMax M3.1 Flash Free", attributes the earlier Space Bunny Free stealth route to this model. OpenRouter's free id minimax/minimax-m3.1-flash:free was created the same UTC day. release_date on this page is 2026-10-19 for the attributed listing; the anonymous Space Bunny Alpha OpenRouter row dates to 2026-09-23.

Compare with

MiniMax M3 Muse Spark Open Muse Wonder Claude Haiku 5.5 any model…

FAQ

Was Space Bunny Alpha actually MiniMax M3.1 Flash?

OpenCode says yes for the free route. models.dev commit fbfb5229 on 2026-10-19 is titled "Reveal Space Bunny as MiniMax M3.1 Flash Free." This site treats that as operator attribution, not a MiniMax press release.

What does MiniMax M3.1 Flash cost?

MiniMax's pricing page lists $0.15 per million input tokens and $0.60 per million output tokens, with cache reads at $0.015 (checked 2026-10-19). That is half of M3's ≤512K standard rate of $0.30/$1.20, and unlike M3 there is no higher tier printed for prompts over 512K.

How does MiniMax M3.1 Flash compare to MiniMax M3 on benchmarks?

On Artificial Analysis HLE without tools, M3.1 Flash is 47.96 against M3's 39.0 — an 8.96-point gap past the 2-point band, a confirmed real lead on that matched independent pair. Terminal-Bench 4.0 on the same runner is 25.76 against M3's 2.02, a 23.74-point gap past the 12.4 band, with no effort tag on either display name. M3's stored cell is vals.ai's 1.01, so the on-site cell comparison is a different harness.

Is MiniMax M3.1 Flash open source?

Not as of 2026-10-19. No MiniMax-M3.1-Flash weights repo showed up in Hugging Face search that day. MiniMax M3 remains the open-weight M-series model this site already tracks.

Which MiniMax M3.1 Flash scores are independent?

Both scores on this page: Artificial Analysis HLE without tools at 47.96 and Terminal-Bench 4.0 at 25.76, read from the AA model-page payload on 2026-10-19. AA-AnalystAgent is null on the MiniMax M3.1 Flash object.

Does MiniMax M3.1 Flash replace MiniMax M3?

No. The previous-version badge needs the same family line and the same tier. MiniMax M3 is the base tier and MiniMax M3.1 Flash is the Flash tier, so M3 does not carry that badge. MiniMax still lists M3 on its pricing page with the 512K length step; both can remain on sale.