Artificial Analysis benchmarks Inkling 2 Small matching Gemini 3.8 Flash performance
11/06/2026 — 11/06, 09:17·4 sources (all)·11 reports (all)
Key takeaways
- According to Artificial Analysis benchmarks, Inkling 2 Small matches the performance of Gemini 3.8 Flash, currently debuting as a proprietar…
Background & analysis
On November 5, 2026, Thinking Machines Lab, led by Mira Murati, officially released its second-generation model, Inkling 2 Small. The model uses a MoE architecture with 594B total parameters and 17B active parameters. According to Artificial Analysis, Inkling 2 Small scores 31 on the Artificial Analysis Intelligence Index. It launched as a proprietary model, available first-party through Thinking Machines' Tinker platform and Tinker Playground, and entered closed testing on day one. The company plans to release the model weights next week after safety testing is complete.
In specific evaluations, Inkling 2 Small scored 1527 Elo on AA-Briefcase, a benchmark that tests long-horizon business workflows involving spreadsheets, presentations, and memos. That is 615 Elo points higher than the previous Inkling Small at 912, and the rubric pass rate reached 2.5x that of its predecessor. On AA-3D, which tests the ability to create interactive 3D experiences, and Code2Video-AA, the model also made large gains over Inkling (xhigh), with scores close to Muse Spark 1.3 and Qwen3.8 Max, respectively. In addition, Artificial Analysis stated that Inkling 2 Small matches Gemini 3.8 Flash (high) in its index, and that once the weights are released it would be the highest-scoring open weights model developed outside China on the current rankings.
The evaluations also noted trade-offs. Inkling 2 Small hallucinates less mainly because it answers fewer questions, while accuracy improves only slightly. On AA-Omniscience v1.1 its net score rose from -9 for Inkling Small to 16, but remains behind Gemini 3.8 Flash (high) at 23 and Claude Sonnet 5.5 (medium) at 19. In terms of usage, the model averages roughly 150K output tokens per Intelligence Index task, about 4x the roughly 37K of Inkling Small, and more than Gemini 3.8 Flash (high) at about 99K and Claude Sonnet 5.5 (medium) at about 31K. The cost per task in the Intelligence Index is $1.18, up nearly 6x from the previous model, with token prices set at $0.30 per 1M input tokens, $1.00 per 1M output tokens, and $0.03 per 1M cache hit tokens. The announcement drew attention in communities such as Reddit's LocalLLaMA, and the story currently stands at the planned open-weights release for next week.
AI-generated from 11 reports · updated 1 hour ago
Latest turnArtificial Analysis reports that Inkling 2 Small matches the performance of Gemini 3.8 Flash (high) on its index. The model is currently launching as a proprietary offering via Thinking Machines' Tinker platform, with open weights expected to make it the top open model outside China.
Related AI tools
- Heat index
- 376
- All sources
- 4
- All reports
- 11
- First seen
- 2 hours ago
Reports on this story headlines open the original
Artificial Analysis reports that Inkling 2 Small matches the performance of Gemini 3.8 Flash (high) on its index. The model is currently launching as a proprietary offering via Thinking Machines' Tinker platform, with open weights expected to make it the top open model outside China.
TechmemeAI score 59The Inkling 2 Small (594B-A17B) model has been announced, with open weights scheduled for release next week. This upcoming open-weight release is drawing attention from local LLM researchers and enthusiasts.
Reddit · LocalLLaMAAI score 59According to community reports, the Inkling 2 Small model has been officially released. This open-weight release has drawn significant attention from the LocalLLaMA community for local deployment.
Reddit · LocalLLaMAAI score 59Thinking Machines Lab, founded by Mira Murati, has released its second-generation model, Inkling 2 Small, featuring 594B total parameters. The model debuted with closed testing and is scheduled for an open-source release next week, achieving an AA Index score of 31.
掘金AI score 68Thinking Machines Lab debuted Inkling 2 Small, an MoE model featuring 594B total parameters and 17B active parameters. The team plans to release the model weights next week following safety testing.
TechmemeAI score 69According to Artificial Analysis, Inkling 2 Small costs $1.18 per task in the Intelligence Index, up nearly 6x from its predecessor. Its token pricing is set at $0.30/$1.00/$0.03 per 1M input/output/cache hit tokens.
X：Artificial Analysis (@artificialanlys)First-partyAI score 54According to Artificial Analysis, Inkling 2 Small averages roughly 150K output tokens per Intelligence Index task, about 4x its predecessor. It also consumes more output tokens than Gemini 3.8 Flash and Claude Sonnet 5.5.
X：Artificial Analysis (@artificialanlys)First-partyAI score 53According to Artificial Analysis, Inkling 2 Small hallucinates less mainly by answering fewer questions, with only slight accuracy gains. Its net score on AA-Omniscience v1.1 reaches 16, trailing Gemini 3.8 Flash and Claude Sonnet 5.5.
X：Artificial Analysis (@artificialanlys)First-partyAI score 53According to Artificial Analysis, Inkling 2 Small makes significant gains over Inkling (xhigh) on AA-3D and Code2Video-AA, scoring close to Muse Spark 1.3 and Qwen3.8 Max respectively.
X：Artificial Analysis (@artificialanlys)First-partyAI score 53According to Artificial Analysis, Inkling 2 Small scores 615 Elo points higher than Inkling Small on AA-Briefcase, achieving 2.5x the rubric pass rate in testing long-horizon business workflows.
X：Artificial Analysis (@artificialanlys)First-partyAI score 53Thinking Machines has released Inkling 2 Small, scoring 31 on the Artificial Analysis Intelligence Index. Its plan to release open weights would make it the highest-scoring open weights model developed outside China.
X：Artificial Analysis (@artificialanlys)First-partyAI score 62
Other stories people are talking about
- Heat index 777SurgeGemini Omni 1.2 Flash arrives on Krea with deeper, cinematic camera control9 sources
- Heat index 76Gemini Canvas turns hand-drawn sketches into working apps1 sources
- Heat index 98NewOpen-source terminal agent Gemini CLI released to help analyze unfamiliar codebases1 sources
- Heat index 60Study finds Gemini and GPT audio models rate Chinese and Nigerian accents lower1 sources
- Heat index 115SurgeGemini 4 Carbon tops Artificial Analysis v5 intelligence index as hallucination rate rises to 11%2 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion