Frontier Log

Benchmark · InactiveNo longer reported by the major labs since Jul 2026

MCP Atlas

Created by Scale AI, 19 Sep 2025 · Higher is better · Website

020406080100Jan 2026Apr 2026Jul 2026Oct 202690.3%Muse Spark 1.285.4%Inkling 2 Small83.6%Gemini 3.5 Flash77.3%Opus 4.788.1%Muse Spark 1.182.2%Muse Spark57.4%Gemini 3 Flash020406080100Jan 2026Apr 2026Jul 2026Oct 202690.3%Muse Spark 1.285.4%Inkling 2 Small83.6%Gemini 3.5 Flash77.3%Opus 4.782.2%Muse Spark

Results

ModelScoreSource
Muse Spark 1.25 Aug 2026 · Meta90.3%Introducing Muse Code and Muse Spark 1.2Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run by Scale AI · the benchmark's own harnessSource image
Muse Spark Open 816B-A42B15 Oct 2026 · Meta89.2%Introducing Muse Spark Open, The Most Capable US Open-Weight Model to DateChecked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run by Scale AI · xhigh effortSource image
Muse Spark 1.19 Jul 2026 · Meta88.1%Introducing Muse Spark 1.1Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run by Scale AISource image
Inkling 2 Small5 Nov 2026 · Thinking Machines85.4%Introducing Inkling 2 SmallCorroborated The same number is printed in 1 other lab document: 1Run by Scale AI · full set; effort 0.99; 1M-token trajectory limitOur capture
Gemini 3.5 Flash19 May 2026 · Google83.6%Gemini 3.5: frontier intelligence with actionCorroborated The same number is printed in 1 other lab document: 1Source image
Muse Spark8 Apr 2026 · Meta82.2%Introducing Muse Spark 1.1Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run by Scale AI · reported laterSource image
Claude Opus 4.716 Apr 2026 · Anthropic77.3%Introducing Claude Opus 4.7Corroborated The same number is printed in 1 other lab document: 1Run by Scale AI · refreshed April 2026 harness; adaptive thinking, max effortSource image
Gemini 3 Flash17 Dec 2025 · Google57.4%Build with Gemini 3 Flash, frontier intelligence that scales with youChecked A person checked this number against its source on 29 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Corroborated The same number is printed in 1 other lab document: 1thinkingSource image

All benchmarks