Xiaomi

MiMo-V2.6-Flash

The value half of Xiaomi's September release: 309B total / 15B activated parameters, MIT weights, 1M context, and API pricing of $0.14/$0.28 per million tokens — below GLM-5.3-Flash's $0.15/$0.50 list rate. Two of its five tracked scores are independent: vals.ai's Terminal-Bench 2.1 run, 76.40, 11.2 points under Xiaomi's 87.6 and, unlike Xiaomi's table, above the Pro's 67.79. The other three come from Xiaomi's model card.

MiMo-V2.6-Flash benchmarks and pricing, every number sourced: 5 tracked MiMo-V2.6-Flash benchmark scores (2 independently run, 3 still resting on a vendor’s own claim), priced at $0.14 per million input tokens and $0.28 per million output.

MiMo-V2.6-Flash architecture: Mixture-of-Experts; 309B total parameters (15B activated per token); 1M-token context window.

Released: 2026-09-22
License: open-weights
Context window: 1M tokens
Knowledge cutoff: Not disclosed
Verified: sources checked 2026-09-22–2026-10-08
Parameters: 309B (15B active)
Architecture: Mixture-of-Experts

MiMo-V2.6-Flash’s verified record

MiMo-V2.6-Flash’s most-compared rival is GLM-5.3-Flash: 5 not callable across their 5 shared comparisons. MiMo-V2.6-Flash is priced at $0.14/$0.28 per 1M tokens (in/out) vs GLM-5.3-Flash’s $0.15/$0.50.

Against the 122 head-to-head comparisons MiMo-V2.6-Flash shares with other tracked models:

5 real gaps11 inside the band106 not called

A gap counts for MiMo-V2.6-Flash only where both sides were run independently and the benchmark still separates models. Losses are listed alongside wins on purpose — MiMo-V2.6-Flash trails on 1 of them.

Terminal-Bench 2.1 Terminal ops

±10.6 is noise

Behind

Claude Opus 5.5 −11.2

Ahead

Tencent Hy4 preview +21.3 · MiniMax M3 +22.8

Tie

8 models within ±10.6

Unverified

3 models — vendor-reported on one side

Setup-dependent

21 models — scored on a different harness or effort tier

Terminal-Bench 4.0 Terminal ops

±12.4 is noise

Ahead

Tencent Hy4 preview +16.2 · MiniMax M3 +23.2

Tie

3 models within ±12.4

Unverified

1 model — vendor-reported on one side

Setup-dependent

4 models — scored on a different harness or effort tier

No verdict for MiMo-V2.6-Flash anywhere on Agents' Last Exam, DeepSWE, Toolathlon-Verified ( nothing independently confirmed on both sides).

None of MiMo-V2.6-Flash’s coding comparisons are independently confirmed on both sides yet.

MiMo-V2.6-Flash API pricing

$0.14 in / $0.28 out per 1M tokens — official pricing source

What MiMo-V2.6-Flash costs per job

MiMo-V2.6-Flash cost for three reference workloads, computed from its list rates
Workload	Tokens in / out	Cost
One long chat turn	100K / 10K	$0.017
A codebase review	1,000K / 100K	$0.168
A day of agent work	10,000K / 1,000K	$1.68

Computed from MiMo-V2.6-Flash’s list rates above — cache discounts and batch tiers are not applied.

MiMo-V2.6-Flash is one of 2 Xiaomi models tracked on this site, at these official list prices.

Xiaomi model pricing, official list rates
Model	In / 1M	Out / 1M
MiMo-V2.6-Flash	$0.14	$0.28
MiMo-V2.6-Pro	$0.43	$0.87

MiMo-V2.6-Flash benchmark scores

MiMo-V2.6-Flash benchmark scores, provenance, and source links
Benchmark	Score
Terminal-Bench 2.1^[1] Terminal ops · ±10.6 is noise	76.4vals.ai · 2026-10-01
DeepSWE^[2] Long-horizon coding · ±9.5 is noise	67.9huggingface.co · 2026-09-22
Toolathlon-Verified^[3] Multi-tool chores · ±9.7 is noise	73.6huggingface.co · 2026-09-22
Agents' Last Exam^[4] Professional work · ±3.2 is noise	27.6huggingface.co · 2026-09-22
Terminal-Bench 4.0^[5] Terminal ops · ±12.4 is noise	24.2vals.ai · 2026-10-08

Who ran these numbers: 2 of 5 independent — vals.ai (2); vendor self-reported (3).

Terminal-Bench 2.1: vals.ai's own Terminus 2 run, board row 'MiMo V2.6 Flash' (76.40 ± 1.72, no effort parameter listed, pass@1), from its archived Terminal-Bench 2.1 table. Replaces Xiaomi's self-reported 87.6 (model-card table, no effort tier stated), 11.2 points higher. Xiaomi's card reports 87.6 — not stored as a row; this cell holds AA's independent run.
DeepSWE: Xiaomi's model-card evaluation table reports 67.9. Xiaomi's announcement instead reports 65.7 as this RL run's after-training endpoint on the same held-out benchmark (up from 48.8 before it) — two official numbers, conflict flagged, neither independently confirmed; deepswe.datacurve.ai did not list this model when checked 2026-09-22.
Toolathlon-Verified: Xiaomi's model-card evaluation table reports 73.6 for the Flash. toolathlon.xyz's board listed only MiMo V2.5 when checked 2026-09-22.
Agents' Last Exam: Xiaomi's model-card evaluation table reports 27.6 for the Flash, split not stated. snorkel.ai's board listed only MiMo V2.5 when checked 2026-09-22.
Terminal-Bench 4.0: vals.ai board row — mini-swe-agent harness (single bash tool), pass@1 averaged over 3 full passes, raw value 24.242 at no effort tier stated. No official or Artificial Analysis Terminal-Bench 4.0 row exists for MiMo-V2.6-Flash. Xiaomi's card reports 28.8 with no tier stated — not stored as a row; this cell holds vals.ai's independent run.

Notes on the record

Pricing checked 2026-09-22 against Xiaomi's official model page (mimo.mi.com/models/en-US/mimo-v2.6-flash): $0.14 per million input tokens, $0.28 output, $0.0028 cached input — one flat rate, no length threshold or promotional condition; Xiaomi states V2.6 pricing matches the V2.5 series (announcement, 2026-09-22).

Spec from the HuggingFace card (XiaomiMiMo/MiMo-V2.6-Flash-RL, MIT license, uploaded 2026-09-21): sparse MoE, 309B total / 15B activated, 1M context, 128K max output, text/image/video/audio input. Knowledge cutoff not disclosed. Same one-day date spread as the Pro: the official page's Update Time reads September 22, Artificial Analysis and the HuggingFace upload say September 21.

Provenance: two of the five tracked scores is independent, the rest are Xiaomi's own model-card figures. The independent one, Terminal-Bench 2.1, is vals.ai's Terminus 2 run, 76.40 (no effort parameter listed), which replaced Xiaomi's 87.6 on 2026-10-01. Unlike Pro, Flash had no Artificial Analysis page on 2026-09-22 (the URL returned 404). Checked live 2026-09-22: tbench.ai, deepswe.datacurve.ai, toolathlon.xyz, snorkel.ai, livebench.ai, arcprize.org and matharena.ai do not list it. The same official-source split on DeepSWE v1.1 applies: the card's eval table says 67.9, while the announcement's training write-up reports 65.7 as this RL run's after-training endpoint (up from 48.8 before it) on the same held-out benchmark; card figure recorded, conflict flagged.

The card's 'OSWorld-Verified' 80.8 states no release version — not recordable under version-sliced OSWorld cells.

Compare with

MiMo-V2.6-Pro Muse Ember Claude Fable 5.2 Inkling 2 Small any model…

FAQ

Is MiMo-V2.6-Flash open source?

Yes. Xiaomi published the weights under an MIT license on HuggingFace (XiaomiMiMo/MiMo-V2.6-Flash-RL) alongside the technical report, RL environments and training framework, announced 2026-09-22.

How much does MiMo-V2.6-Flash cost to run?

$0.14 per million input tokens and $0.28 per million output tokens on Xiaomi's official pricing page (checked 2026-09-22), with cached input at $0.0028. One flat rate — no length threshold, time-of-day discount or promotion — and Xiaomi says it matches the V2.5-series price.

Has MiMo-V2.6-Flash been independently benchmarked?

On two benchmarks. vals.ai's archived Terminal-Bench 2.1 board lists it at 76.40, against Xiaomi's own 87.6, and that run now replaces the vendor figure here; vals.ai's Terminal-Bench 4.0 run (24.24, default effort) joined on 2026-10-08. On 2026-09-22 Artificial Analysis had no page for it (404), and none of tbench.ai, deepswe.datacurve.ai, toolathlon.xyz, snorkel.ai, livebench.ai, arcprize.org or matharena.ai listed it. The other three tracked scores come from Xiaomi's own model card.

How does MiMo-V2.6-Flash compare to MiMo-V2.6-Pro?

Same license, context window and modalities in a smaller package: 309B/15B parameters versus 1.02T/42B, at roughly a third of the input price. On Xiaomi's own table it trails the Pro on all four shared benchmarks, for example Terminal-Bench 2.1 at 87.6 versus 89.9. vals.ai's independent Terminal-Bench 2.1 runs reverse that order, Flash 76.40 against Pro 67.79, though the 8.6-point gap sits inside the 10.6-point noise band.

What is MiMo-V2.6-Flash's best verified score?

Terminal-Bench 2.1 at 76.4 (2026-10-01). As of the October 2026 refresh it is the only independently-run figure on this sheet — everything else here traces back to Xiaomi's own launch materials, which is why the verified column is the one to watch.

What context window does MiMo-V2.6-Flash support?

1,048,576 tokens — the MiB-scale window shared across the MiMo-V2.6 pair, matching MiMo-V2.6-Pro's figure. No length-based price tier applies on this model.