Model Markets
Get Free API KeyAPI

Thinking Machines LabInkling

External Link

Inkling is Thinking Machines Lab's open-weight multimodal mixture-of-experts model with 975B total parameters, 41B active parameters, native text, image, audio, and video input, and a 1M-token context window.

Token context
1,049K tokens
Inputs
Text, Image, Video, Audio
Outputs
Text
Released
Jul 15, 2026

Benchmark Market Position

Intelligence, Cost, and Efficiency

All rankings →

Intelligence

Rank #27 of 57

  1. 100.0*Claude Opus 5.5
  2. 99.0Claude Fable 5.1
  3. 90.9*GPT-5.5 Pro
  4. 90.0Gemini 3.8 Flash
  5. 85.9GPT-6.1 Sol
  6. 84.6GPT-6 Astra
  7. 83.4*Qwen3.8 Max
  8. 83.3*DeepSeek V4.1 Flash
  9. 82.1Gemini 3.1 Pro
  10. 77.3GPT-5.6 Terra
  11. 74.0*Grok 4.7
  12. 73.5*ERNIE 5.1
  13. 73.3Kimi K3
  14. 71.9*GLM 5.3
  15. 67.5*Qwen3.8 Flash
  16. 64.3*Grok 4.20 Multi Agent
  17. 63.3*Qwen3.7 Plus
  18. 63.1*Claude Sonnet 5
  19. 57.9*DeepSeek V4 Pro
  20. 56.2*GLM 5.3 Flash
  21. 54.6*Mistral Large 4
  22. 54.3*Gemma 4 31B
  23. 52.2Qwen3.8 27B
  24. 50.5*Qwen3.5 397B A17B
  25. 50.1*Hy3
  26. 47.4*Gemma 4 26B A4B
  27. 47.4Inkling
  28. 43.0*Mistral Large 3
  29. 42.3GPT-6 Luna
  30. 37.8*NVIDIA Nemotron 3 Ultra 550B A55B BF16
  31. 35.2*Inkling Small
  32. 33.9*MiniMax M3
  33. 27.0*Kimi K2.7 Code
  34. 25.1*Grok Build 0.1
  35. 20.8GPT-5.4 Mini
  36. 17.8Gemini 3.5 Flash Lite
  37. 17.2*Qwen3.5 35B A3B
  38. 15.7GPT-5.4 Nano
  39. 12.8*Muse Glimmer 30B
  40. 11.7*Solar Pro 4
  41. 9.3*Claude Haiku 4.5
  42. 0.4*DeepSeek R1
  43. 0.0*Olmo 3 32B Think
  44. 0.0*Nova 2 Lite
  45. 0.0*Command A
  46. 0.0*Granite 4.2 30B
  47. 0.0*Granite 4.2 3B
  48. 0.0*Granite 4.2 8B
  49. 0.0*Llama 4 Maverick 17B 128E Instruct
  50. 0.0*Llama 4 Scout 17B 16E Instruct
  51. 0.0*phi-4
  52. 0.0*NVIDIA Nemotron 3 Nano 30B A3B BF16
  53. 0.0*NVIDIA Nemotron 3.5 Lightning 30B A3B
  54. 0.0*GPT-4o
  55. 0.0*GPT-4o Mini
  56. 0.0*gpt-oss-120b
  57. 0.0*gpt-oss-20b
* Provisional: fewer than 3 sources
View full ranking →

Cost

Rank #13 of 27

  1. $0.0020Grok Build 0.1
  2. $0.0087GLM 5.3 Flash
  3. $0.017MiniMax M3
  4. $0.018GPT-6 Luna
  5. $0.021Qwen3.8 Flash
  6. $0.022DeepSeek V4.1 Flash
  7. $0.029Gemini 3.5 Flash Lite
  8. $0.044Kimi K2.7 Code
  9. $0.054Mistral Large 4
  10. $0.054DeepSeek V4 Pro
  11. $0.058GPT-5.4 Nano
  12. $0.063Qwen3.8 27B
  13. $0.090Inkling
  14. $0.092NVIDIA Nemotron 3 Ultra 550B A55B BF16
  15. $0.104GPT-6.1 Sol
  16. $0.107Qwen3.8 Max
  17. $0.160Gemini 3.8 Flash
  18. $0.161Gemini 3.1 Pro
  19. $0.163Kimi K3
  20. $0.194Claude Sonnet 5
  21. $0.214GPT-5.4 Mini
  22. $0.248GLM 5.3
  23. $0.266GPT-5.6 Terra
  24. $0.341Grok 4.7
  25. $0.512Claude Opus 5.5
  26. $0.520GPT-6 Astra
  27. $0.824Claude Fable 5.1
View full ranking →

Efficiency

Rank #22 of 27

  1. 71.7*DeepSeek V4.1 Flash
  2. 65.8*GLM 5.3 Flash
  3. 64.2*Qwen3.8 Flash
  4. 62.5*Grok Build 0.1
  5. 60.1GPT-6.1 Sol
  6. 58.6*Qwen3.8 Max
  7. 58.5Gemini 3.8 Flash
  8. 54.6Gemini 3.1 Pro
  9. 53.9*Claude Opus 5.5
  10. 52.8GPT-6 Luna
  11. 51.5*DeepSeek V4 Pro
  12. 50.1Kimi K3
  13. 49.9*Mistral Large 4
  14. 49.5Claude Fable 5.1
  15. 49.0*MiniMax M3
  16. 48.0GPT-5.6 Terra
  17. 47.3Qwen3.8 27B
  18. 46.1GPT-6 Astra
  19. 45.9*GLM 5.3
  20. 44.3*Grok 4.7
  21. 43.5*Claude Sonnet 5
  22. 42.0Inkling
  23. 37.8*Kimi K2.7 Code
  24. 37.1*NVIDIA Nemotron 3 Ultra 550B A55B BF16
  25. 36.6Gemini 3.5 Flash Lite
  26. 29.9GPT-5.4 Nano
  27. 21.6GPT-5.4 Mini
* Provisional: fewer than 3 sources
View full ranking →

The orange outline marks Inkling. Each available panel preserves its real rank and value; missing benchmark or pricing inputs remain explicitly unranked.

Providers & Pricing

Estimate workload cost →
ProviderInput / 1MOutput / 1MCache read / 1MCache write / 1MObservedEvidence
$0.95$4.05$0.16—Oct 15, 2026Provider-reported
$1.00$4.05$0.17—Oct 19, 2026Provider-reported

Benchmarks

All benchmark results →
ARC-AGI-1verified-v1-613c385b0c9379.50percent——InklingunknownOriginal sourcewinner eligibleThird-party benchmark
ARC-AGI-2verified-v2-be40992860f936.53percent——InklingunknownOriginal sourcewinner eligibleThird-party benchmark
LiveBench2026-06-25-90f8d44da96576.47percent22,2131,270inkling-xhighaverage per caseRecomputedwinner eligibleThird-party benchmark
LMArena Agent Arenaagent-2026-10-16-a64ca3b12b6e-14.35score—43,507Inkling; 95% CI [-15.81859883, -12.88168979]; sessions 43507; observations 1662352; rank 50unknownOriginal sourcewinner eligibleThird-party benchmark
LMArena Text Arenatext-2026-10-15-a64ca3b12b6e1,441.38rating—40,648inkling; 95% CI [1437.08506747, 1445.68368363]; votes 40648; rank 74unknownOriginal sourcewinner eligibleThird-party benchmark
ToneBench2026-10-17-10-task-e9c76358afb268.77points6,99850Inkling (high)average per caseOriginal sourcenot winner eligibleThird-party benchmark

At a Glance

Model Facts

Identity

Status
active
Developer ID
thinkingmachines/Inkling
Released
Jul 15, 2026
Version
Inkling
License
apache-2.0

Capacity

Token context
1,049K tokens
Maximum output
Unknown
Total parameters
975B
Active parameters
41B
Knowledge cutoff
Not reported

Interface and Access

Inputs
Text, Image, Video, Audio
Outputs
Text
API available
Yes
Open weights
Yes
Self-hostable
Yes
Reasoning
Yes
Vision
Yes
Tool calling
Yes
Capabilities
agents, chat, coding, generation, reasoning, tools, vision

This page represents one developer model product. Dated API snapshots, serving endpoints, pricing, regions, quantizations, and service tiers attach as versioned aliases or provider details and never create duplicate public model records.

Version Lineage

PredecessorUnknown
SuccessorsUnknown
Aliasesinkling, thinkingmachines/Inkling, Thinking Machines Inkling

Explicit Unknowns

  • Maximum output
  • Knowledge cutoff

Primary Evidence

Sources and Observation Date

thinkingmachines/Inkling developer model repositoryThinking Machines Lab · observed Sep 3, 2026thinkingmachines/Inkling README.md at 828496eeae4cThinking Machines Lab · observed Sep 3, 2026Inkling: Our Open-Weights ModelThinking Machines Lab · observed Sep 3, 2026thinkingmachines/Inkling config.json at 828496eeae4cThinking Machines Lab · observed Sep 3, 2026thinkingmachines/Inkling immutable Hub API metadataThinking Machines Lab · observed Sep 3, 2026Inkling Model CardThinking Machines Lab · observed Sep 3, 2026thinkingmachines/Inkling tokenizer_config.json at 828496eeae4cThinking Machines Lab · observed Sep 3, 2026Deepinfra catalogDeepinfra · observed Oct 19, 2026Fireworks Ai catalogFireworks Ai · observed Oct 19, 2026Openrouter catalogOpenrouter · observed Oct 19, 2026Together Ai catalogTogether Ai · observed Oct 19, 2026

Comparable Peers

Related Model Comparisons

All comparisons →
APairBContext
vscross-developer peersimage, text
vscross-developer peersimage, text
vscross-developer peersimage, text, video
vscross-developer peerstext
vscross-developer peerstext
vscross-developer peersaudio, image, text, video
vscross-developer peersaudio, image, text
vscross-developer peersaudio, image, text, video
vscross-developer peerstext
vsfamily variantsaudio, image, text

Questions

Inkling FAQs

What is Inkling?+

Inkling is Thinking Machines Lab's open-weight multimodal mixture-of-experts model with 975B total parameters, 41B active parameters, native text, image, audio, and video input, and a 1M-token context window. It is developed by Thinking Machines Lab and its current sourced lifecycle status is active.

What inputs and outputs does Inkling support?+

Inkling's current record lists Text, Image, Video, and Audio as input and Text as output.

How much context does Inkling support?+

The current record lists a 1,049K-token context window and does not report a maximum output length.

Is Inkling available through an API?+

Inkling is marked as API-available. The directly sourced provider records currently include Deepinfra, Fireworks Ai, Openrouter, and Together Ai.

How large is Inkling?+

The current record lists 975B total parameters and 41B active parameters. Its architecture is InklingForConditionalGeneration.

What capabilities and tasks does Inkling support?+

Inkling's recorded capabilities are agents, chat, coding, generation, reasoning, tools, and vision. Its supported tasks are audio, coding, image, reasoning, text, and video.

How much does Inkling cost through an API?+

The lowest directly sourced prices currently attached to Inkling are $0.95 per million input tokens through Deepinfra and $4.05 per million output tokens through Deepinfra. Prices are provider-specific and should be checked against each cited observation date.

Can Inkling be self-hosted?+

Inkling is an open-weight model. Self-hosting is supported, and the recorded license is apache-2.0.

Does Inkling have related versions or aliases?+

No predecessor is recorded. No successor is recorded. Recorded aliases are inkling, thinkingmachines/Inkling, and Thinking Machines Inkling.

Which models can Inkling be compared with?+

The current compatible peer set includes DeepSeek V4.1 Flash, Kimi K3, Nova 2 Lite, NVIDIA Nemotron 3 Ultra 550B A55B BF16, and DeepSeek V4 Pro Base. Compatibility requires shared sourced modalities and tasks; it does not imply that one model is better. Browse model comparisons →

Send Feedback