Thinking Machines' Inkling 2 Small scores 31 on Artificial Analysis Intelligence Index, matching Gemini 3.8 Flash (high)
Thinking Machines has released Inkling 2 Small, which scores 31 on the Artificial Analysis Intelligence Index v5.0.1, matching Gemini 3.8 Flash (high). It launches as a proprietary model on Tinker at $0.30/$1.00 per 1M input/output tokens, with an Apache 2.0 weights release planned.
Inkling 2 Small at a glance
How Inkling 2 Small ranks now
| Benchmark | Score | Rank | Source |
|---|---|---|---|
| AA Intelligence Index | 30.6 index | 43 of 124 | Artificial Analysis, Nov 2026 ↗ |
| Speed (tok/s) | 106 tok/s | 60 of 127 | Artificial Analysis, Nov 2026 ↗ |
Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.
Thinking Machines has released Inkling 2 Small, which scores 31 on the Artificial Analysis Intelligence Index v5.0.1. That score is up 18 points from Inkling Small and matches Google DeepMind's Gemini 3.8 Flash (high). It also surpasses Xiaomi's MiMo-V2.6-Pro. According to Artificial Analysis, if Thinking Machines follows through on its plan to release the weights, Inkling 2 Small would be the highest-scoring open weights model developed outside China based on current rankings.
Model details
- Type: Proprietary reasoning model at launch. An Apache 2.0 weights release is planned, with the date unconfirmed.
- Size: 594B total parameters, 17B active.
- Modalities: Text and image input, text output.
- Pricing: $0.30 per 1M input tokens, $1.00 per 1M output tokens, $0.03 per 1M cached input tokens.
- Availability: Thinking Machines' Tinker platform and Tinker Playground.
Benchmark results (Artificial Analysis)
Gains over Inkling Small are largest on agentic tasks:
- AutomationBench-AA: 46%, up from 5%.
- FrontierCode 2.0: 25%, up from 5%. Claude Sonnet 5.5 (medium) scores 11%.
- Terminal-Bench 4.0: 13%, against 30% for Claude Sonnet 5.5 (medium).
- HLE-Diamond: 29%, up from 12%.
- CritPt v2 (research-level physics): 56%, up from 34%, but behind Gemini 3.8 Flash (high) at 65%.
- AA-Briefcase (long-horizon business workflows): 1527 Elo vs. 912 for Inkling Small, 1439 for Claude Sonnet 5.5 (medium) and 1203 for Gemini 3.8 Flash (high). Inkling 2 Small passes 50% of rubric checks vs. 20% for Inkling Small.
- AA-3D: 1021 Elo vs. 966 for Gemini 3.8 Flash and 1033 for Meta's Muse Spark 1.3 (max). Inkling (xhigh) scored 583.
- Code2Video-AA: 1020 Elo vs. 899 for Gemini 3.8 Flash and 1021 for Alibaba's Qwen3.8 Max (0902) (xhigh). Inkling (xhigh) scored 467.
Artificial Analysis also reports that Inkling 2 Small scores above Claude Haiku 5.5 and Grok 4.7 on AA-3D, and behind Claude Sonnet 5.5 on both AA-3D and Code2Video-AA.
Hallucinations and token use
On AA-Omniscience v1.1, the hallucination rate falls to 27% from Inkling Small's 61%. Accuracy rises only slightly, to 34% from 32%, while the attempt rate drops to 53% from 77%. Artificial Analysis attributes the improvement mainly to the model answering fewer questions. Net score improves to 16 from -9, still behind Gemini 3.8 Flash (high) at 23 and Claude Sonnet 5.5 (medium) at 19.
The model averages about 150K output tokens per Intelligence Index task. Inkling Small uses about 37K, Claude Sonnet 5.5 (medium) about 31K, and Gemini 3.8 Flash (high) about 99K.
Cost per Intelligence Index task is $1.18, up almost 6x from Inkling Small's $0.20. Gemini 3.8 Flash (high) costs $7.77 per task and Claude Sonnet 5.5 (medium) costs $10.02.
What this means
Inkling 2 Small's per-task cost is low for its score because its token prices are low. That holds even though it generates far more output tokens than the comparison models. Teams comparing it with Gemini 3.8 Flash (high) or Claude Sonnet 5.5 (medium) should weigh task-level cost and latency from long outputs, not only per-token prices.
The coding results are split. It leads Claude Sonnet 5.5 (medium) on FrontierCode 2.0 but trails it on Terminal-Bench 4.0, so results will depend on the workload.
The lower hallucination rate comes from selective answering, not higher accuracy, so the model declines more questions. The open-weights ranking is conditional: it applies only once Thinking Machines publishes the weights, and no date has been confirmed.
Related Articles
Thinking Machines Lab launches Inkling 2 Small, a 594B-parameter sparse MoE model, on Tinker
Thinking Machines Lab introduced Inkling 2 Small on November 5, 2026. The sparse Mixture-of-Experts model has 594 billion total parameters and 17 billion active parameters, and is available through Tinker with closed weights at launch.
ByteDance Seed releases Cola2, a 9B continuous-latent language model with 60.10 eight-task average
ByteDance Seed has published Cola2, a 9B-parameter base language model that generates continuous latent blocks with a diffusion transformer and realizes them as discrete tokens. Trained on 346.03B source tokens, it reports a 60.10 average across eight tasks. Weights are on Hugging Face; training code is not included.
Preferred Networks opens PLaMo 3 Translate weights in four sizes, 0.8B to 31B
Preferred Networks released PLaMo 3 Translate, a translation-specialized LLM, as open weights on November 5 in four sizes: 0.8B, 2B, 8B and 31B. The company claims world-class translation performance and says even the smallest model outperforms Google's TranslateGemma and Tencent's Hy-MT2.
Salesforce launches Koa Decisions, a Gemma-based model that classifies text, image and audio requests
Salesforce introduced Koa Decisions on November 6, 2026. The decision model is built from Google's open-weight Gemma and accepts text, images and audio. It is the next version of HyperClassifier.
Comments
Loading...