Wait Which Model?
← Back to directory
Ant Group (InclusionAI)

Ling-2.6-1T

Released Apr 24, 2026 · knowledge cutoff unpublished

Status
Superseded
Location
China
Modality
Text
Context window
262K
Max output
33K
Speed
Price ($/MTok in / out)
$0.30 / $2.50
Cost per task
Open weights
Yes
Benchmarks4 of 8 reported

GPQA Diamond

75.2%

SWE-bench Verified

72.2%

AIME (latest)

87.4%

Humanity's Last Exam

8.2%

MMLU-Pro

Terminal-Bench 2.1

LMArena Elo

ARC-AGI-2

Strengths & weaknesses

Strengths

  • Skips multi-step chain-of-thought in favor of a "fast thinking" mode, giving high-quality answers at roughly a quarter the token cost of comparable reasoning models
  • Hybrid MLA/linear-attention architecture keeps latency and VRAM use down at the full 256K context length
  • Open weights under a permissive licence, with drop-in compatibility for existing agent scaffolds (Claude Code, OpenCode, and others)

Weaknesses

  • A non-reasoning model by design — not built for deep multi-step reasoning tasks, where its Ring-2.6 thinking sibling is the intended tool instead
  • Humanity's Last Exam score is low relative to its coding and agentic numbers, consistent with a model tuned for execution over broad expert knowledge
  • Too new for independent coverage on LMArena or Artificial Analysis' full benchmark suite at launch

Ant Group's Bailing team announced Ling-2.6-1T on 2026-04-24 (weights open-sourced 2026-04-30 on Hugging Face/ModelScope); Maia's supplied link (ling-1t.ai/ling-2-6-1t) is not an official Ant Group/InclusionAI domain and was not used as a source. GPQA Diamond, HLE and AIME26 figures are third-party aggregator reports (aiflashreport.com, howaiworks.ai), not confirmed on an official Ant Group table or on Artificial Analysis directly — treat as secondary-sourced. Pricing is Artificial Analysis' median across API providers ($0.30/$2.50 per MTok); OpenRouter separately lists a lower $0.075/$0.625 rate, which may be promotional. Context window is 262,144 tokens via the public API; some coverage cites a 1M-token native/trained context. Total/active parameters are reported as roughly 1.03T/63B (MoE).

For developers

API model strings

Not researched

Licence

MIT License · Permissive · commercial use permitted

Retirement

No retirement announced

Lineage

No recorded predecessor

News