Inkling 2 Small

Multimodal
Thinking Machines Lab

Inkling 2 Small is Thinking Machines Lab's second-generation MoE multimodal model (594B total / 17B active parameters), released on 2026-11-05. It accepts text and image input (no audio) and generates text with variable thinking effort. The model card lists a context window up to 1M tokens, while Tinker serves 256K for serverless inference and 64K/256K for training and sampling. It scores 82.2% on SWE-Bench Verified, 93.0% on GPQA Diamond and 44.1% on Humanity's Last Exam, with a vendor-reported Artificial Analysis Intelligence Index v5.0 of 30.6. Weights are closed for now with an Apache 2.0 release pending external safety testing; Tinker serverless pricing is $0.30 / $1.00 per 1M input/output tokens.

Key Specifications

Parameters
594.0B
Context
256.0K
Release Date
November 5, 2026
Average Score
67.1%

Timeline

Key dates in the model's history
Announcement
November 5, 2026
Last Update
November 9, 2026
Today
November 20, 2026

Technical Specifications

Parameters
594.0B
Training Tokens
-
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.30
Output (per 1M tokens)
$1.00
Max Input Tokens
256.0K
Max Output Tokens
-
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests
SWE-Bench Verified
SWE-Bench Verified. Vendor model card table; harness not stated (Inkling-Small's 80.2% used a bash-only harness, so not like-for-like). • Self-reported
82.2%

Reasoning

Logical reasoning and analysis
GPQA
GPQA Diamond. Vendor model card table. • Self-reported
93.0%

Other Tests

Specialized benchmarks
AA-Briefcase
AA-Briefcase v1.1 Elo. Vendor launch post / model card. • Self-reported
50.9%
Artificial Analysis
Artificial Analysis Intelligence Index v5.0. Score 30.6 (vendor model card table; the official launch post rounds it to 31). • Self-reported
30.6%
BrowseComp
BrowseComp with context management. Vendor model card. • Self-reported
90.2%
CharXiv-R
Charxiv RQ original (without Python). Vendor model card (original/with Python 85.6/87.7). • Self-reported
85.6%
GDPval-AA
GDPVal-AA v2.1 Elo. Vendor model card table. • Self-reported
54.3%
Humanity's Last Exam
HLE (text only). Vendor model card table. HLE-Diamond separately reported at 28.7%. • Self-reported
44.1%
MCP Atlas
MCP Atlas Public set. Vendor model card (Full/Public 85.4/85.0). • Self-reported
85.0%
MMMU-Pro
MMMU Pro (Standard 10). Vendor model card table. • Self-reported
81.6%
Tau3 Banking
Tau 3 Banking v1.0.1. Vendor model card table. • Self-reported
40.4%

License & Metadata

License
proprietary
Announcement Date
November 5, 2026
Last Updated
November 9, 2026

Compare Inkling 2 Small

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.