Skip to content

Gyroscape models / Introducing

GyroThink1.5 Pro

Our frontier coding model.

For the code that asks more of you. Bring complex reasoning, a large working context, and development tools together in Gyroscape.

Token context window
1 million
Maximum output tokens
131,072
With native tool use
Text + vision

Built for substantial work

Keep the whole problem in view.

Reason across the codebase.

Bring requirements, relevant files, and previous decisions into the same working context. A 1 million-token window gives substantial projects room to breathe.

Move from explanation to implementation.

Work with Gyroscape’s file, terminal, and development tools to inspect a problem, make changes, and review the result in the context of your project.

Bring cybersecurity into the build.

Use Gyroscape’s Cyber capability to review code for security issues, investigate findings, and carry fixes through testing. Keep the evidence and remediation alongside the implementation.

See the interface. Understand the task.

Native image input lets you bring screenshots and visual references into the conversation. Text, images, and tool results can inform the same coding task.

Explore Cyber

Performance / Reference

Coding, in perspective.

Published coding comparisons, including Kimi K3, GLM-5.3, and Claude Opus 5. Higher scores are better.

DeepSWE v1.1
GyroThink 1.5 Pro71.5
  1. Claude Opus 574.0
  2. GPT-5.6 Sol73.0
  3. Kimi K369.0
  4. GLM-5.369.0
Terminal-Bench 2.1
GyroThink 1.5 Pro82.3
  1. GPT-5.6 Sol85.8
  2. Claude Opus 584.6
  3. Kimi K380.9
  4. GLM-5.371.5
FrontierSWE v2
GyroThink 1.5 Pro30.0
  1. Claude Opus 552.0
  2. GPT-5.6 Sol32.2
  3. GLM-5.330.2
  4. Kimi K325.9
ProgramBench
GyroThink 1.5 Pro21.2
  1. Claude Opus 537.0
  2. GPT-5.6 Sol23.0
  3. GLM-5.319.0
  4. Kimi K317.5
SWE-Marathon v1.1
GyroThink 1.5 Pro40.1
  1. Claude Opus 550.0
  2. Kimi K348.1
  3. GPT-5.6 Sol42.5
  4. GLM-5.342.5
Terminal-Bench 3.0
GyroThink 1.5 Pro31.9
  1. Claude Opus 542.7
  2. GPT-5.6 Sol34.6
  3. GLM-5.332.4
  4. Kimi K317.4

Published results / September 14, 2026

The latest frontier comparisons.

GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5, with the benchmark versions preserved.

Swipe the table to compare all models →

BenchmarkGyroThink 1.5 ProGPT-6 AstraClaude Fable 5.1Claude Opus 5
Terminal-Bench 4.035.257.955.852.6
DeepSWE v1.171.274.167.473.7
FrontierCode 1.1 Extended58.964.563.663.6

The details

Before you start building.

What is GyroThink 1.5 Pro?

GyroThink 1.5 Pro is Gyroscape’s frontier coding model for reasoning through complex codebases, working with development tools, supporting cybersecurity reviews, and carrying implementation tasks through verification.

Is GyroThink 1.5 Pro fine-tuned on the Kimi K3 base model?

Yes. GyroThink 1.5 Pro is fine-tuned from the Kimi K3 base model for Gyroscape’s coding and cybersecurity workflows.

What are the context window and maximum output limits?

GyroThink 1.5 Pro has a 1 million-token context window and a maximum output of 131,072 tokens per response. The context window is the working space for the conversation, code, and tool results; it is separate from the maximum length of a generated response. Request and account limits still apply.

Can GyroThink 1.5 Pro help with cybersecurity work?

Yes. Use it with Gyroscape’s Cyber capability to review your code for security issues, investigate findings, propose fixes, and verify changes with the project’s tools. Findings and remediation stay connected to the same development task.

How should I read the GyroThink benchmark figures?

Higher percentages indicate stronger performance on the named benchmark. Use the benchmark names and versions shown when comparing results.

From first idea to final diff

Bring a harder problem.

Open your project in Gyroscape and select GyroThink 1.5 Pro when it appears in your account’s model picker.