Skip to main content

Model profile

Thinking Machines Inkling model

Data verified
Overall Score
69Prov. #30 of 78Verified #17 of 32
Arena Elo
Not listed
Categories Ranked
3of 8
Price (1M tokens)
$1.87 in / $4.68 out
API pricing
Speed
Not listed
Context
1M

Evidence coverage

16 of 300 tracked benchmarks are published. 16 are verified and 0 provisional. 6 of 8 categories are measured.

Updated July 15, 2026Methodology
Published / tracked
16 / 300
Verified
16
Provisional
0
Categories measured
6 / 8

Evidence by category

  • Agentic4 benchmarks
    Verified
  • Coding3 benchmarks
    Verified
  • Reasoning0 benchmarks
    Not measured
  • Knowledge4 benchmarks
    Verified
  • Math1 benchmark
    Verified
  • Multilingual0 benchmarks
    Not measured
  • Multimodal3 benchmarks
    Verified
  • Inst. Following1 benchmark
    Verified
Open WeightSelf-hostHybrid
Confidence:
High
base

According to BenchLM.ai, Inkling ranks #30 out of 78 models on the provisional leaderboard with an overall score of 69/100. It also ranks #17 out of 32 on the verified leaderboard. This places it in the mid-tier of AI models, with strengths in specific benchmark categories.

Inkling is a open weight model with a 1M token context window. It processes queries without explicit chain-of-thought reasoning, offering faster response times and lower token usage.

This profile currently has 16 of 300 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Its strongest category is Instruction Following (#19), while its weakest is Multimodal & Grounded (#42). This performance profile makes it a well-rounded choice across a range of tasks.

Peer position

Exact provisional scores and ranks for the closest listed peers.

Range 68.070.0

  1. DeepSeek V4 Flash (Max)
    DeepSeek
    #2870.0
    DeepSeek V4 Flash (Max) is #28 with a score of 70.0.
    Compare
  2. Kimi K2.5 (Reasoning)
    Moonshot AI
    #2970.0
    Kimi K2.5 (Reasoning) is #29 with a score of 70.0.
    Compare
  3. InklingCurrent model
    Thinking Machines Lab
    #3069.0
    Inkling is #30 with a score of 69.0.
  4. Grok 4.20
    xAI
    #3168.0
    Grok 4.20 is #31 with a score of 68.0.
    Compare
  5. Claude Opus 4.7
    Anthropic
    Unranked69.0
    Claude Opus 4.7 is Unranked with a score of 69.0.
    Compare
  6. Qwen3.5 397B (Reasoning)
    Alibaba
    Unranked69.0
    Qwen3.5 397B (Reasoning) is Unranked with a score of 69.0.
    Compare
  7. Grok 4.1 Fast
    xAI
    Unranked68.0
    Grok 4.1 Fast is Unranked with a score of 68.0.
    Compare

Category percentile

More

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

  1. Inst. Following87%
    Eligible cohort rank #19 of 135Category score 84.9
  2. Agentic71%
    Eligible cohort rank #36 of 121Category score 66.8
  3. Multimodal63%
    Eligible cohort rank #42 of 111Category score 63.8

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #36 of 121Percentile 71stWeight 22%4 benchmarksVerified66.8
CodingRank Not rankedWeight 20%3 benchmarksVerified73.0
ReasoningWeight 17%0 benchmarksNot measuredNot measured
KnowledgeRank Not rankedWeight 12%4 benchmarksVerified74.9
MathRank Not rankedWeight 5%1 benchmarkVerified78.9
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalRank #42 of 111Percentile 63rdWeight 12%3 benchmarksVerified63.8
Inst. FollowingRank #19 of 135Percentile 87thWeight 5%1 benchmarkVerified84.9

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Agentic4 benchmarks
Terminal-Bench 2.0Provider exact
63.8%Weighted 38%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 63.8% on Terminal-Bench 2.1 with the best harness. The source assigns zero to rollouts whose web search reveals solution contamination; BenchLM stores the result in the existing Terminal-Bench 2 slot until a separate 2.1 key exists.
BrowseCompProvider exact
77.1%Weighted 28%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 77.1% on BrowseComp with context management.
MCP AtlasProvider exact
74.1%Display only
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 74.1% on MCP Atlas.
Design Arena Agentic Web DevBenchmark exact

Design Arena Agentic Web Dev Elo

1258Display only
Source: Design Arena: Agentic Web Dev leaderboardProvenance: The live Design Arena API reports model ID camellia at 1258 Elo from 1,152 blinded battles. The official Design Arena registry maps camellia to the display name Inkling and provider Thinking Machines. BenchLM stores the current Elo as display-only.
Coding3 benchmarks
SWE-bench VerifiedProvider exact

Software Engineering Benchmark Verified

77.6%Weighted 16%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 77.6% on SWE-bench Verified.
SWE-bench ProProvider exact
54.3%Weighted 10%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 54.3% on the public SWE-bench Pro split.
Terminal-Bench 2.0Provider exact
63.8%Display only
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 63.8% on Terminal-Bench 2.1 with the best harness. The source assigns zero to rollouts whose web search reveals solution contamination; BenchLM stores the result in the existing Terminal-Bench 2 slot until a separate 2.1 key exists.
Knowledge4 benchmarks
HLEProvider exact

Humanity's Last Exam

46%Weighted 32%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 46.0% on HLE with tools.
GPQAProvider exact

Graduate-Level Google-Proof Q&A

87.9%Weighted 5%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 87.9% on GPQA Diamond. BenchLM stores the exact result on both the weighted GPQA lane and the GPQA Diamond display key.
GPQA-DProvider exact

GPQA Diamond

87.9%Display only
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 87.9% on GPQA Diamond.
HLE w/o toolsProvider exact

Humanity's Last Exam without tools

30%Display only
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 30.0% on text-only HLE.
Math1 benchmark
AIME26Provider exact

AIME 2026

97.1%Weighted 25%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 97.1% on AIME 2026.
Multimodal3 benchmarks
MMMU-ProProvider exact

Massive Multi-discipline Multimodal Understanding Pro

73.5%Weighted 45%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 73.5% on MMMU Pro Standard 10.
CharXivProvider exact

CharXiv Reasoning

82%Weighted 25%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 82.0% on CharXiv RQ with Python.
CharXiv w/o toolsProvider exact

CharXiv Reasoning without tools

78.1%Display only
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 78.1% on CharXiv RQ without Python.
Inst. Following1 benchmark
IFBenchProvider exact

Instruction Following Benchmark

79.8%Weighted 65%
Source: Thinking Machines Lab: Inkling launch postProvenance: Thinking Machines reports Inkling at effort=0.99 scoring 79.8% on IFBench.

Frequently Asked Questions

How does Inkling perform overall in AI benchmarks?

Inkling currently ranks #30 out of 78 models on BenchLM's provisional leaderboard with an overall score of 69. It also ranks #17 out of 32 on the verified leaderboard. It is created by Thinking Machines Lab and features a 1M context window.

Is Inkling good for knowledge and understanding?

Inkling has visible benchmark coverage in knowledge and understanding, but BenchLM does not currently assign it a global category rank there.

Is Inkling good for coding and programming?

Inkling has visible benchmark coverage in coding and programming, but BenchLM does not currently assign it a global category rank there.

Is Inkling good for mathematics?

Inkling has visible benchmark coverage in mathematics, but BenchLM does not currently assign it a global category rank there.

Is Inkling good for agentic tool use and computer tasks?

Inkling ranks #36 out of 78 models in agentic tool use and computer tasks benchmarks with an average score of 66.8. There are stronger options in this category.

Is Inkling good for multimodal and grounded tasks?

Inkling ranks #42 out of 78 models in multimodal and grounded tasks benchmarks with an average score of 63.8. There are stronger options in this category.

Is Inkling good for instruction following?

Inkling ranks #19 out of 78 models in instruction following benchmarks with an average score of 84.9. There are stronger options in this category.

Is Inkling open source?

Yes, Inkling is an open weight model created by Thinking Machines Lab, meaning it can be downloaded and run locally or fine-tuned for specific use cases.

Does Inkling have full benchmark coverage on BenchLM?

Not yet. Inkling currently has 16 published benchmark scores out of the 300 benchmarks BenchLM tracks. BenchLM only exposes non-generated public benchmark rows, so missing categories stay blank until a sourced evaluation is available.

What is the context window size of Inkling?

Inkling has a context window of 1M, which determines how much text it can process in a single interaction.

Don't miss the next GPT moment

Which models moved up, what is new, and what it costs. One email each week.

Free. One email per week.