Skip to content

Meta Proprietary Released Sep 2, 2026 Updated

Muse Spark 1.3

ToolsGuide Scores, rankings, test results and price.

Model summary

  • Text #16 / 416

    585 / 1,000

    ToolsGuide Score· votes only

  • Vision #20 / 167

    563 / 1,000

    ToolsGuide Score· votes only

  • Coding #21 / 147

    482 / 1,000

    ToolsGuide Score· votes 507, tests 447

  • Agents #10 / 53

    520 / 1,000

    ToolsGuide Score· votes 490, tests 558

  • Price per 1M tokens

    In $0.10· Out $0.20

    $0.10

    Input · median text model $0.43

Summary

Muse Spark 1.3 is Meta's model, released Sep 2, 2026. It's ranked in 4 categories and does best at text: #16 of 416, with a ToolsGuide Score of 585 against a median of 235.

It also ranks #20 of 167 for vision (563), #21 of 147 for coding (482), #10 of 53 for agents (520).

Its text and vision score uses votes only, because it has published fewer than two of the tests used for those categories.

It costs $0.10 per million input tokens and $0.20 per million output tokens, below the median text model ($0.43).

It has published results on 11 tests and has the top score on 5 of them, including MRCR v2 (8-needle) and Job Bench.

Specifications

Creator
Meta
Licence
Proprietary
Released
Sep 2, 2026
Context window
1M tokens
Input price
$0.10 / 1M
Output price
$0.20 / 1M
Tests reported
11
Settings ranked
Muse Spark 1.3 (Max), Muse Spark 1.3 (xHigh)

Model comparison

Against the ten best text models.

Text score

ToolsGuide Score out of 1,000 · Higher is better

Vision score

ToolsGuide Score out of 1,000 · Higher is better

Input price

USD per 1M input tokens · Lower is better

Leaderboards

Where Muse Spark 1.3 sits among the top 20 in each category.

Text

#16 of 416 · ToolsGuide Score out of 1,000, where 1,000 is a flawless model

Full text leaderboard →

Vision

#20 of 167 · ToolsGuide Score out of 1,000, where 1,000 is a flawless model

Full vision leaderboard →

Coding

#21 of 147 · ToolsGuide Score out of 1,000, where 1,000 is a flawless model

Full coding leaderboard →

Agents

#10 of 53 · ToolsGuide Score out of 1,000, where 1,000 is a flawless model

Full agents leaderboard →

By task

Scored from votes on each kind of prompt alone, on the same 1,000-point scale.

Score against price

Text · the 10 priced models ranked closest · up and to the left is better value

Against its nearest rivals

Text prompts by kind · ToolsGuide Score from votes · hover for values

Text

Muse Spark 1.3 (Max) · hover a bar for its rank

Vision

Muse Spark 1.3 (Max) · hover a bar for its rank

Coding

Muse Spark 1.3 (Max) · hover a bar for its rank

Test results

Every test Muse Spark 1.3 has a published result on, out of 100%, with its rank among the models that took it.

  • DeepSWE 1.1

    #7 / 50

    DeepSWE 1.1 evaluates software engineering agents using the mini-swe-agent harness.

    75.4%

    Self-reported · In the coding score

  • Terminal-Bench 2.1

    #4 / 46

    Terminal-Bench 2.1 is an updated release of the Terminal-Bench benchmark that tests AI agents' ability to operate a computer via the terminal. It evaluates how well models handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, data science workflows, and security tasks.

    88.8%

    Self-reported

  • MRCR v2 (8-needle)

    #1 / 32

    MRCR v2 (8-needle) is a variant of the Multi-Round Coreference Resolution benchmark that includes 8 needle items to retrieve from long contexts. This tests models' ability to simultaneously track and reason about multiple pieces of information across extended conversations.

    98.5%

    Self-reported · In the text score

  • AutomationBench

    #5 / 24

    AutomationBench is a tool-use benchmark that evaluates AI agents on automating real-world workflows, testing their ability to orchestrate tools and complete multi-step automation tasks.

    49.4%

    Self-reported · In the agents score

  • OSWorld 2.0

    #7 / 17

    OSWorld 2.0 is a benchmark of 108 long-horizon, real-world computer-use workflows spanning everyday and professional tasks. Each task is an end-to-end workflow that takes human users a median of about 1.6 hours, scored with a binary-completion metric, and targets challenges such as dynamic environments, cross-source reasoning, and implicit-state inference.

    66.9%

    Self-reported

  • GDPval-AA

    #4 / 14

    GDPval-AA evaluates AI agents on economically valuable professional knowledge-work tasks and reports performance as an Elo score.

    58.5%

    Self-reported

  • DeepSearchQA

    #5 / 10

    DeepSearchQA is a benchmark for evaluating deep search and question-answering capabilities, testing models' ability to perform multi-hop reasoning and information retrieval across complex knowledge domains.

    89.4%

    Self-reported

  • Job Bench

    #1 / 9

    Job Bench evaluates AI agents on realistic professional tasks that require multi-step planning, research, and production of work artifacts.

    64.9%

    Self-reported

  • SWE Atlas - Codebase QnA

    #1 / 5

    SWE Atlas - Codebase QnA evaluates a model's ability to answer questions about real codebases, measuring repository-level comprehension and the ability to reason about code structure, behavior, and intent across an entire project.

    59.4%

    Self-reported

  • MRCR v2 (8-needle, 512K-1M)

    #1 / 4

    MRCR v2 8-needle variant evaluated on contexts from 512K to 1M tokens.

    98.1%

    Self-reported

  • Agentic IF Index (Internal)

    #1 / 1

    Meta's internal Agentic Instruction-Following Index aggregates evaluations of compliance with tool-use constraints, fine-grained rubrics, compound constraints, long policies, and instructions embedded in structured workflows.

    57.8%

    Unverified