Skip to content

Inkling-Small

Thinking Machines LabOpen weightsHybridVerifiedReleased Jul 30, 2026Input: text, image, audio
Compare with InklingSource scores ↗OpenRouter ↗
Overall score
52.7
Rank 89 of 222 · 44.3–61.1
Context window
1M
1,000,000 tokens
Price per 1M tokens
$0.58 in · $1.44 out
Source: BenchLM
Blended price
$0.80
3:1 input:output tokens
Output speed
—
Not measured yet

Category scores

Scores per capability area with the model's rank among ranked models in that category.

  1. 1Instruction following#993.3
  2. 2Math#569.8
  3. 3Knowledge#6368.4
  4. 4Multimodal#3767.5
  5. 5Agentic & tool use63.1
  6. 6Coding#7257.0
  7. 7Reasoning55.0
Higher is better. #N is the model's rank in that category, where it is eligible to be ranked.

Benchmark results (20)

As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.

Agentic & tool use

Terminal-Bench 2.164.7#48 of 68
BrowseComp77.4#25 of 50
MCP Atlas79.6#12 of 42
Toolathlon-Verified54.4#19 of 22

Math

FrontierMath v2 (Tiers 1-3)46.316#43 of 64
FrontierMath v2 (Tier 4)17.073#38 of 51
AIME2695.5#10 of 29
HMMT Feb 202690.2#8 of 24

Coding

SWE-bench Verified80.2#17 of 98
SWE-bench Pro55.9#45 of 69
SciCode48.7#7 of 23

Multimodal

MMMU-Pro74.0#35 of 48
CharXiv81.3#23 of 42
CharXiv w/o tools77.4#13 of 14

Knowledge

GPQA89.5#29 of 117
HLE47.8#20 of 69
HLE w/o tools31.6#29 of 43

Reasoning

ARC-AGI-240.1#36 of 41
ARC-AGI-184.00#31 of 36

Instruction following

IFBench82.2#3 of 42

Pricing by source

USD per 1M tokens.

SourceInputOutputCached in
List price (BenchLM)$0.58$1.44$0.12
OpenRouter thinkingmachines/inkling-small$0.45$1.20$0.10

Thinking Machines Lab's Tinker pricing docs list the 64K Inkling-Small tier at $0.58 prefill / $0.116 cached prefill / $1.44 sample per million tokens during a limited-time 50% discount. The hosted 256K fine-tuning tier costs $1.16 / $0.232 cached / $2.89, while the beta 256K serverless endpoint costs $0.30 / $0.06 cached / $1.20. We store the lower-context fine-tuning tier as the headline Tinker rate and the open checkpoint's 1M-token ceiling separately.

Arena ratings

Human-preference ratings from LMArena. Best-rated variant shown.

Text1413#135 · 32.3K votes
WebDev (coding)1409#91 · 5.3K votes
Vision1237#75 · 8.1K votes

Open weights

From the Hugging Face repository.

Parameters
266B
Licence
apache-2.0
Downloads (30 days)
786.2K

Speed

Measured by us through OpenRouter (median of trials).

Not measured yet.

Related models

Same family

Similar score