MiniMax-M2.5

MiniMaxAI MiniMax (Modified MIT) (modified-mit) out 8 months ago

229B parameters (8 of 256 experts per token), text in, text out, 192K tokens of context. 8 providers serve it from $0.27 per million input tokens.

Verified 2026-10-25 Rank #42 in LLMs · score 109.1 method v1 Sources 4 Downloads 30d 992.9k · #109 of 619 tracked

Can I run it?137GB at INT4Estimatefits 2×80 GBmethod

What does it cost?$0.27/1M incheapest of 7Venice2026-10-24

Can I use it?MiniMax (Modified MIT) (modified-mit)MIT with added terms; read the repo LICENSE before commercial hostinglicence file2026-10-24

Does it work?75.8on SWE-bench Verified#1 of 22 measuredleaderboard2026-02-17

Do people use it?992.9kdownloads, 30 days#109 of 619 we track · top 18%+5 likes, downloads +28% this week1,526 likesHugging Face2026-10-24

Worth reading

Picked by rule from 20 tracked pieces: the lab's own post, the most shared, a field report, the latest.

MiniMax M2.5 is trained by Claude Opus 4.6 according to a user observation in the discussion.

From MiniMax M2.5 is trained by Claude Opus 4.6? · news.ycombinator.com · 2026-03-14

The lab's own post

MiniMax M2.5 released: 80.2% in SWE-bench Verified

MiniMax releases M2.5 and M2.5-Lightning, two versions identical in capability but differing in speed.

minimax.io 2026-02-12 207 HN pts · 55 comments 2.2k words sells inference

Most discussed · 12 HN pts

MiniMax M2.5 is trained by Claude Opus 4.6?

Anthropic identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude's capabilities.

MiniMax M2.5 is trained by Claude Opus 4.6 according to a user observation in the discussion.
Gemini 2.0 Exp 1206 was reported to be indirectly trained on Claude's outputs with humans in between.

news.ycombinator.com 2026-03-14 12 HN pts · 11 comments 643 words

Latest · 8 months ago

MiniMax M2.5 SOTA in Coding and Agent, Designed for Agent Universe

MiniMax M2.5 is available in 100 TPS and 50 TPS versions with output pricing at just 1/10 to 1/20 of comparable models.

minimax.io 2026-02-14 2 HN pts 344 words sells inference

20 best of 20 tracked

The 4 picks above, plus these 16. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleMiniMax M2.5agent.minimax.io · 2026-02-11 · 55 HN pts · official
ArticleMiniMax M2.5 is beating Claude Opus 4.6 and MiniMax is 17x-20x cheaperswebench.com · 2026-03-02 · 6 HN pts
ArticleMiniMax M2.5: Built for Real-World Productivity. - MiniMax News | MiniMaxminimax.io · 2026-02-12 · official
ArticleMiniMax-M2.5: How to Run Guideunsloth.ai · 2026-02-23 · 3 HN pts
ArticleMiniMax M2.5: Faster, stronger, smarter. Built for real-world productivityminimaxi.com · 2026-02-12 · 4 HN pts · official
DiscussionThe RL Architecture Behind Minimax M2.5twitter.com · 2026-02-21 · 3 HN pts
ArticleMiniMax M2.5 matches Claude Opus at 1/33rd the costnews.reading.sh · 2026-02-12
ArticleMiniMax M2.5 Review: Frontier Code at Bargain Cost | Awesome Agentsawesomeagents.ai · 2026-03-09
ArticleWhat is MiniMax M2.5?apidog.com · 2026-03-03
ArticleMiniMax M2.5 Is Goodziva.sh · 2026-02-19
ArticleMiniMax M2.5 | Awesome Agentsawesomeagents.ai · 2026-02-24
ArticleMiniMax M2.5 codes on a top level without the costi-scoop.eu · 2026-02-14
ArticleMiniMax-M2.5: The $1/hour Frontier Model – Maxime Labonnemlabonne.github.io · 2026-02-13
ArticleMiniMax M2.5: Best 8-Benchmark Review, $1/Hour Pricingbinaryverseai.com · 2026-02-13
ArticleMiniMax M2.5 Analysis: The New Frontier in Coding & Function Callingvertu.com · 2026-02-13
ArticleMiniMax M2.5: Frontier Coding for Less | S5 Labss5labs.io · 2026-02-12

Show all 16

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

7 providers from $0.27

7 providers with a live price, from $0.27 to $0.30 per million input tokens; the median is $0.30. One dot per provider, cheapest in teal.

#	Cheapest 5	In /1M	Out /1M	Context	Quant	Seen
1	Venice	$0.27	$0.95	193K	unknown	Oct 24
2	StreamLake	$0.27	$1.08	195K	unknown	Oct 24
3	AtlasCloud	$0.29	$1.20	192K	fp8	Oct 24
4	DigitalOcean	$0.30	$1.20	64K	unknown	Oct 24
5	GMICloud	$0.30	$1.20	192K	fp8	Oct 24

All 8 providers →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Widely served (8 API providers); heavily run locally (63 quantized builds).

As a service8AtlasCloud, DigitalOcean, Featherless, GMICloud +4
On own hardware63quantized builds
Built on3027 fine-tunes · 1 adapters · 2 merges
In apps100+Hugging Face Spaces

Top Spaces running it: akhaliq/anycoder ♥ 3,321 · hadadxyz/ai ♥ 819 · pliny-the-prompter/obliteratus ♥ 621 · GenAISecurityProject/OWASP-AIBOM-Generator ♥ 30

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits two 80 GB GPUs at INT4; too big for one

About 137 GB at INT4, 274 GB at FP8, 549 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

Estimate params × bytes per param × 1.2 · how we estimate

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#1 of 22 on SWE-bench Verified so far

75.8. Each tick is a model on this site with a measured SWE-bench Verified row; author-reported numbers stay in Reference. Percent resolved. agent: mini-SWE-agent, reasoning_effort=high

Source · leaderboard · 2026-02-17

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2026-02-12), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.

What people built on it

93 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

93 derivatives on the Hub, most of them quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF unsloth/MiniMax-M2.5-GGUF for llama.cpp, Ollama, LM Studio 310.3k downloads MLX lmstudio-community/MiniMax-M2.5-MLX-8bit for Apple silicon 383.6k downloads NVFP4 nvidia/MiniMax-M2.5-NVFP4 for vLLM, SGLang 1.1M downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

5 rows

Benchmarks 5 rows

measured third-party leaderboard row (2)author-reported number from the model card or eval files (3)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
measured SWE-bench Multilingual · Write code with agents	68.3 percent resolved	agent: mini-SWE-agent	leaderboard	2026-02-16
measured SWE-bench Verified · Write code with agents	75.8 percent resolved	agent: mini-SWE-agent, reasoning_effort=high	leaderboard	2026-02-17
author-reported diamond (Idavidrein/gpqa)	85.2	—	Model Card	2026-02-13
author-reported hle (cais/hle)	19.4	—	Model Card	2026-02-13
author-reported swe_bench_%_resolved (SWE-bench/SWE-bench_Verified)	75.8	high reasoning	SWE-Bench official evaluation	2026-10-24

Use-case evidence

5 use cases

What it is good at 5 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in LLMs · MoE

Facts

MiniMax-M2.5 has 229B parameters. (safetensors metadata, observed 2026-10-24)
MiniMax-M2.5 is a mixture-of-experts model with 256 experts, routing 8 per token. (config.json, observed 2026-10-24)
Context length is 196,608 tokens (192K). (config.json max_position_embeddings, observed 2026-10-24)
License: MiniMax (Modified MIT) (modified-mit) (friction: review). MIT with added terms; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
Created 2026-02-12, last modified 2026-03-10. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	MiniMaxM2ForCausalLM
Model type	minimax_m2
Params by dtype	BF16: 1,230,021,632 · F32: 48,774,656 · F8_E4M3: 227,410,968,576
Size band	XXL
MoE	yes
Experts	256
Experts per token	8
Context length	196,608 (config.json max_position_embeddings)
License	MiniMax (Modified MIT) (modified-mit) (other)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · MiniMaxM2ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · MiniMaxM2ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for MiniMax-M2.5: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	549 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)	274 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	137 GB · Multi-GPU node (<640 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

40 variants

Lineage

Original release (shipped quantized by the lab) Family: MiniMaxAI/MiniMax-M2.5 93 derivatives on the Hub

quantized: 63finetune: 27merge: 2adapter: 1

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
nvidia/MiniMax-M2.5-NVFP4	quantized	NVFP4	116B	XXL	other	nvidia	41	1.1M
lmstudio-community/MiniMax-M2.5-MLX-8bit	quantized	MLX	229B	XXL	other	lmstudio-community	3	383.6k
unsloth/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	other	unsloth	247	310.3k
lukealonso/MiniMax-M2.5-NVFP4	quantized	NVFP4	116B	XXL	mit	lukealonso	45	309.1k
lmstudio-community/MiniMax-M2.5-MLX-4bit	quantized	MLX	229B	XXL	other	lmstudio-community	1	303.2k
lmstudio-community/MiniMax-M2.5-MLX-6bit	quantized	MLX	229B	XXL	other	lmstudio-community	0	298.6k
QuantTrio/MiniMax-M2.5-AWQ	quantized	AWQ	229B	XXL	other	QuantTrio	15	193.8k
mratsim/MiniMax-M2.5-BF16-INT4-AWQ	quantized	-BF16	39.1B	XL	other	mratsim	43	168.6k
cyankiwi/MiniMax-M2.5-AWQ-4bit	quantized	AWQ	36.8B	XL	other	cyankiwi	11	97.8k
ubergarm/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	—	ubergarm	51	40.4k
amd/MiniMax-M2.5-MXFP4	quantized	MXFP4	116B	XXL	other	amd	1	38.5k
lmstudio-community/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	other	lmstudio-community	5	31.7k
mratsim/MiniMax-M2.5-FP8-INT4-AWQ	quantized	-FP8	39.1B	XL	other	mratsim	21	26.2k
mradermacher/MiniMax-M2.5-i1-GGUF	quantized	GGUF	—	—	other	mradermacher	1	25.4k
togethercomputer/Aurora-Spec-Minimax-M2.5	finetune	—	858M	S	apache-2.0	togethercomputer	6	18.9k
thoughtworks/MiniMax-M2.5-Eagle3	finetune	—	243M	S	apache-2.0	thoughtworks	7	18.3k
mudler/MiniMax-M2.5-APEX-GGUF	quantized	GGUF	—	—	other	mudler	3	16.9k
Ex0bit/MiniMax-M2.5-PRISM-LITE	finetune	GGUF	—	—	other	Ex0bit	3	15.6k
ox-ox/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	mit	ox-ox	19	14.1k
cerebras/MiniMax-M2.5-REAP-139B-A10B	quantized	—	139B	XXL	other	cerebras	29	9.8k
dealignai/MiniMax-M2.5-UNCENSORED-JANG_2L	finetune	—	60.2B	XL	apache-2.0	dealignai	11	9.4k
Intel/MiniMax-M2.5-int4-AutoRound	quantized	int4	31.7B	L	—	Intel	13	8.8k
DevQuasar/MiniMaxAI.MiniMax-M2.5-GGUF	quantized	GGUF	—	—	—	DevQuasar	7	8.1k
mlx-community/MiniMax-M2.5-8bit	quantized	-8bit	229B	XXL	other	mlx-community	2	8k
QuantPasture/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	—	QuantPasture	30	6.5k
mlx-community/MiniMax-M2.5-4bit	quantized	-4bit	229B	XXL	other	mlx-community	3	7.6k
mlx-community/MiniMax-M2.5-3bit	quantized	—	229B	XXL	other	mlx-community	11	7k
exdysa/MiniMax-M2.5-REAP-172B-A10B-GGUF-Q4_K_M	quantized	GGUF	—	—	other	exdysa	3	6.9k
zenlm/zen5-mini-gguf	quantized	gguf	—	—	other	zenlm	0	5k
Ex0bit/MiniMax-M2.5-PRISM-PRO	finetune	GGUF	—	—	other	Ex0bit	18	3.9k
mlx-community/MiniMax-M2.5-6bit	quantized	—	229B	XXL	other	mlx-community	4	4.4k
cerebras/MiniMax-M2.5-REAP-172B-A10B	quantized	—	173B	XXL	other	cerebras	27	2.3k
mradermacher/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	other	mradermacher	0	3.7k
selimaktas/MiniMax-M2.75-460B-A20B	merge	—	453B	XXL	other	selimaktas	27	2k
unsloth/MiniMax-M2.5	quantized	—	229B	XXL	other	unsloth	5	3.1k
wimmmm/MiniMax-M2.5-REAP-172B-A10B-GGUF	quantized	GGUF	—	—	other	wimmmm	2	3.2k
dealignai/MiniMax-M2.5-JANG_3L-CRACK	finetune	—	116B	XXL	apache-2.0	dealignai	3	2.5k
marksverdhei/MiniMax-M2.5-GGUF	quantized	GGUF	—	—	other	marksverdhei	3	2.4k
mlx-community/MiniMax-M2.5-5bit	quantized	—	229B	XXL	other	mlx-community	1	2.4k
EliasOenal/MiniMax-M2.5-Hybrid-AWQ-W4A16G128-Attn-fp8_e4m3-KV-fp8_e4m3	quantized	AWQ	33.8B	L	other	EliasOenal	16	1.2k

Repository changelog

17 commits

Repository changelog 17 commits

Date	Commit	SHA
2026-03-10	Add evaluation results for GPQA, HLE (#3)	f710177
2026-03-10	Add evaluation results on SWE-Bench Verified (#42)	1825d90
2026-03-06	Create LICENSE-MODEL	ab0ddff
2026-02-16	Update README.md	1248334
2026-02-14	Update README.md	37649bd
2026-02-13	Update README.md	3040bea
2026-02-13	Update README.md	2663734
2026-02-13	Update README.md	3e16da2
2026-02-13	Upload bench_12.png	c6c3633
2026-02-13	update: tool calling guide	5fb9455
2026-02-12	update configs	134ae7c
2026-02-12	Add files using upload-large-folder tool	9ba43d5
2026-02-12	Add files using upload-large-folder tool	afa2694
2026-02-12	Add files using upload-large-folder tool	8c436e5
2026-02-12	update docs	6ff2856
2026-02-12	Upload 13 files	aa487ef
2026-02-12	initial commit	e838ad0

All providers

8 providers · 8 endpoints

Where to run 8 providers · 8 endpoints

Hosted endpoints observed for MiniMax-M2.5, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.27/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Venice openrouter	minimax/minimax-m2.5	193K	unknown	$0.27	$0.95	—	—	yes	live	2026-10-24
StreamLake openrouter	minimax/minimax-m2.5	195K	unknown	$0.27	$1.08	—	—	yes	live	2026-10-24
AtlasCloud openrouter	minimax/minimax-m2.5	192K	fp8	$0.29	$1.20	—	—	yes	live	2026-10-24
Novita hf, openrouter	minimax/minimax-m2.5	200K	fp8	$0.30	$1.20	42.5	783 ms	yes	live	2026-10-25
GMICloud openrouter	minimax/minimax-m2.5	192K	fp8	$0.30	$1.20	—	—	yes	live	2026-10-24
DigitalOcean openrouter	minimax/minimax-m2.5	64K	unknown	$0.30	$1.20	—	—	yes	live	2026-10-24
Minimax openrouter	minimax/minimax-m2.5	200K	fp8	$0.30	$1.20	—	—	yes	live	2026-10-24
Featherless hf	MiniMaxAI/MiniMax-M2.5	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

20 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official5

MiniMax M2.5 released: 80.2% in SWE-bench Verified

minimax.io2026-02-12officialauthorityHN front pageHN 207 pts · 55 comments2,175 words

MiniMax releases M2.5 and M2.5-Lightning, two versions identical in capability but differing in speed. M2.5 operates at 50 tokens per second, whereas M2.5-Lightning achieves a steady throughput of 100 tokens per second. The faster M2.5-Lightning version costs $0.3 per million input tokens and $2.4 per million output tokens.written, reviewed, cited

MiniMax M2.5

agent.minimax.io2026-02-11officialauthorityHN 55 pts · 4 comments176 words

MiniMax M3.1 Flash is a new multimodal coding model built for real work tasks. From now through October 25, 2026, 23:59 (UTC+8), all subscribers can use the model with no usage caps. From today through October 31, 2026, 23:59 (UTC+8), new subscribers and M Plan upgraders get 50% off the first month.written, reviewed, cited

MiniMax M2.5 SOTA in Coding and Agent, Designed for Agent Universe

minimax.io2026-02-14officialauthorityHN 2 pts · 0 comments344 words

MiniMax M2.5 is available in 100 TPS and 50 TPS versions with output pricing at just 1/10 to 1/20 of comparable models. Model weights are fully open-sourced on HuggingFace and GitHub for local deployment, supporting vLLM or SGLang for optimal performance. Two API versions exist: M2.5 and M2.5-highspeed, which offer identical results but faster speed.written, reviewed, cited

MiniMax M2.5: Built for Real-World Productivity. - MiniMax News | MiniMax

minimax.io2026-02-12officialauthority2,175 words

MiniMax M2.5: Faster, stronger, smarter. Built for real-world productivity

minimaxi.com2026-02-12officialHN 4 pts · 0 comments522 words

Analysis3

MiniMax-M2.5 - Intelligence, Performance & Price Analysis

artificialanalysis.aiauthority887 words

MiniMax-M2.5 is a deprecated open weights model released on February 12, 2026, featuring 230 billion total parameters with only 10 billion active during inference. The model operates under an MIT license, supports a 205k token context window, and generates text at 70 tokens per second. It scores 23 on the Artificial Analysis Intelligence Index while costing $0.30 per 1M input tokens.written, reviewed, cited

The model is deprecated and benchmarking continues only for the default 10k input token workload.
For a blended rate using a 7:2:1 cache hit/input/output ratio, the cost is $0.20 per 1M tokens.

MiniMax M2.5 is beating Claude Opus 4.6 and MiniMax is 17x-20x cheaper

swebench.com2026-03-02HN 6 pts · 9 comments202 words

MiniMax-M2.5: How to Run Guide

unsloth.ai2026-02-23authorityHN 3 pts · 0 comments921 words

Community2

MiniMax M2.5 is trained by Claude Opus 4.6?

news.ycombinator.com2026-03-14HN 12 pts · 11 comments643 words

Anthropic identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude's capabilities. These labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts, violating terms of service and regional access restrictions. Unambiguous distillation is identified when a model repeats guardrail prompt injections from a teacher model, word-by-word, such as GLM-4.7 repeating injections from Gemini 3.0 Pro.written, reviewed, cited

MiniMax M2.5 is trained by Claude Opus 4.6 according to a user observation in the discussion.
Gemini 2.0 Exp 1206 was reported to be indirectly trained on Claude's outputs with humans in between.

The RL Architecture Behind Minimax M2.5

twitter.com2026-02-21HN 3 pts · 0 comments

All coverage tracked for this model (20) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
MiniMax-M2.5 - Intelligence, Performance & Price Analysis	artificialanalysis.ai	—	2026-10-07	exa
MiniMax M2.5 is trained by Claude Opus 4.6?	news.ycombinator.com	2026-03-14	2026-10-07	hn
MiniMax M2.5 Review: Frontier Code at Bargain Cost \| Awesome Agents	awesomeagents.ai	2026-03-09	2026-10-07	exa
What is MiniMax M2.5?	apidog.com	2026-03-03	2026-10-07	exa
MiniMax M2.5 is beating Claude Opus 4.6 and MiniMax is 17x-20x cheaper	swebench.com	2026-03-02	2026-10-07	hn
MiniMax M2.5 \| Awesome Agents	awesomeagents.ai	2026-02-24	2026-10-07	exa
MiniMax-M2.5: How to Run Guide	unsloth.ai	2026-02-23	2026-10-07	hn
The RL Architecture Behind Minimax M2.5	twitter.com	2026-02-21	2026-10-07	hn
MiniMax M2.5 Is Good	ziva.sh	2026-02-19	2026-10-07	hn
MiniMax M2.5 SOTA in Coding and Agent, Designed for Agent Universe	minimax.io official	2026-02-14	2026-10-07	hn
MiniMax M2.5 codes on a top level without the cost	i-scoop.eu	2026-02-14	2026-10-07	exa
MiniMax-M2.5: The $1/hour Frontier Model – Maxime Labonne	mlabonne.github.io	2026-02-13	2026-10-07	exa
MiniMax M2.5: Best 8-Benchmark Review, $1/Hour Pricing	binaryverseai.com	2026-02-13	2026-10-07	exa
MiniMax M2.5 Analysis: The New Frontier in Coding & Function Calling	vertu.com	2026-02-13	2026-10-07	exa
MiniMax M2.5 matches Claude Opus at 1/33rd the cost	news.reading.sh	2026-02-12	2026-10-07	hn
MiniMax M2.5 released: 80.2% in SWE-bench Verified	minimax.io official	2026-02-12	2026-10-07	hn
MiniMax M2.5: Faster, stronger, smarter. Built for real-world productivity	minimaxi.com official	2026-02-12	2026-10-07	hn
MiniMax M2.5: Built for Real-World Productivity. - MiniMax News \| MiniMax	minimax.io official	2026-02-12	2026-10-07	exa
MiniMax M2.5: Frontier Coding for Less \| S5 Labs	s5labs.io	2026-02-12	2026-10-07	exa
MiniMax M2.5	agent.minimax.io official	2026-02-11	2026-10-07	hn

Similar models

3 models

Similar models

MiniMax-M2.7 · 229B
same size band (XXL), also MoE, same task (text-generation)
Qwen3-235B-A22B · 235B
same size band (XXL), also MoE, same task (text-generation), permissive license
MiniMax-M2 · 229B
same size band (XXL), also MoE, same task (text-generation), fewer providers

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:33 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:00 UTC · Last verified 2026-10-25

Follow MiniMax-M2.5

Watch this model

/feeds/models/MiniMaxAI--MiniMax-M2.5.xml(RSS 2.0) · every change to MiniMax-M2.5
/feeds/families/MiniMaxAI--MiniMax-M2.5.xml(RSS 2.0) · every change across the MiniMax-M2.5 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.