Gemini 4 Carbon Hits Devin Desktop, CLI

// 3h agoMODEL RELEASE

Gemini 4 Carbon Hits Devin Desktop, CLI

Cognition has made Google’s Gemini 4 Carbon available in Devin Desktop and Devin CLI, where it scores 40.0% on FrontierCode 2.0 versus 26.3% for Gemini 4 Argon. The model targets long-horizon coding, debugging, and complex agentic workflows.

// ANALYSIS

This is a substantial upgrade for Devin users, though the benchmark measures the combined model-and-agent harness rather than model quality alone.

–Carbon delivers a 13.7-point improvement over Argon on Cognition’s 132-task coding benchmark
–Its $4.04 average cost per rollout is lower than Argon’s $6.51 on the current leaderboard
–FrontierCode evaluates mergeability using tests, rubrics, and verifiers designed by open-source maintainers
–Carbon’s deeper reasoning makes it better suited to architecture, debugging, and multi-step migrations than routine edits
–Desktop and CLI availability gives developers a practical way to test Carbon inside existing Devin workflows

// TAGS

gemini-4-carbonllmcoding-agentai-codingreasoningclidevtool

DISCOVERED

3h ago

2026-11-10

PUBLISHED

4h ago

2026-11-10

RELEVANCE

10/ 10

AUTHOR

devindevelopers

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

RESEARCH49m ago

Meta's AutoRefine Calibrates Code Judges

Meta’s AutoRefine paper presents a prompt-rewriting loop that uses noisy workflow outcomes—tests, merges, and abandons—to recalibrate LLM code judges, while abstaining when outcomes cannot be explained by the judge’s inputs. On held-out diffs from a deployed coding agent, agreement rose from Scott’s π 0.16 to 0.51. ([paper](https://arxiv.org/abs/2611.09291))

INFRA55m ago

Browser Use Cloud Clears CAPTCHA Bottleneck

Browser Use’s stealth cloud browser combines hardened Chromium, residential proxies, and automatic CAPTCHA solving for AI agents. The company reports 90.5% successful visits across 100 bot-protected sites in its own benchmark.

RESEARCH1h ago

Sycophantic AI Warnings Miss Their Mark

A preregistered study of 2,610 people found that warnings reduced perceived AI objectivity and trust, but did not reliably reduce users’ belief that they were right or their willingness to repair interpersonal conflicts. [Read the paper](https://arxiv.org/abs/2606.21317)