Gemini 3.5 FlashvsComposer 2.5
Gemini 3.5 Flash | Composer 2.5 | |
|---|---|---|
| Benchmarks | ||
Nonsense detection BullshitBench v2Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. | 20% | — |
Prompt injection robustness Gray Swan IPI · k = 1Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. Gray Swan's indirect prompt injection benchmark measures how often such an attack succeeds when the attacker gets a single try. Lower is better. | 14.1% | — |
Prompt injection robustness Gray Swan IPI · k = 10Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 10 tries and counts an attack as successful if any of them works. Lower is better. | 54.2% | — |
Prompt injection robustness Gray Swan IPI · k = 15Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 15 tries and counts an attack as successful if any of them works. Lower is better. | 60.5% | — |
Agentic coding SWE-Bench ProCan the AI fix real bugs in real software? It's handed actual problems from open-source projects and has to write code that genuinely solves them. Higher is better. | 55.1%Best | 54% |
Multilingual coding SWE-Bench MultilingualLike SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. | — | 79.8% |
Agentic coding CursorBench v3.2Cursor's own test of harder, real-world coding tasks inside a code editor, on the refreshed v3.2 task set. Scores aren't comparable with v3.1. Higher is better. | 48.8% | 56.1%Best |
Agentic coding CursorBench v3.1Cursor's own test of harder, real-world coding tasks inside a code editor. Higher is better. | 49.8% | 63.2%Best |
Agentic coding DeepSWE 1.1Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. (Version 1.1 of the test.) Higher is better. | 37% | — |
Agentic coding DeepSWE 1.0Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. Higher is better. | — | 18% |
ML engineering MLE-BenchCan the AI do the work of a machine-learning engineer? It competes in real Kaggle competitions — building, training, and tuning models end to end — and the score reflects how well it places. Higher is better. | 49.7% | — |
Next.js coding Next.js EvalsVercel's open eval of how well AI coding agents build and migrate real Next.js apps — measured as the share of tasks the agent completes successfully. Higher is better. | — | 92% |
Agentic terminal coding Terminal-Bench 2.1Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better. | 76.2%Best | 73% |
Agentic terminal coding Terminal-Bench 2.0Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? (Version 2.0 of the test.) Higher is better. | — | 69.3% |
Multi-step tool use MCP AtlasCan the AI chain together many tools and steps to complete one bigger task, rather than doing just a single thing? Higher is better. | 83.6% | — |
General tool use ToolathlonTests how well the AI uses everyday real-world tools and apps to get things done. Higher is better. | 56.5% | — |
Browser agent BU BenchCan the AI drive a real web browser to finish tasks — clicking, filling forms, and navigating sites the way a person would? Run by Browser Use on their BU Bench task set. Higher is better. | 58% | — |
Multidisciplinary reasoning Humanity's Last Exam · no toolsHumanity's Last Exam — extremely hard expert questions across many subjects, written so you can't just look up the answer. “No tools” means the AI answers on its own. Higher is better. | 40.2% | — |
Abstract reasoning ARC-AGI-2Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better. | 72.1% | — |
Agentic computer use OSWorld-VerifiedCan the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better. | 78.4% | — |
Agentic financial analysis Finance Agent v2Tests the AI on real financial-analysis work, like digging through reports and making sound decisions. Higher is better. | 57.9% | — |
Knowledge work GDPval-AAMeasures how well the AI does economically valuable knowledge work, judged against human experts. Shown as a rating (like a chess Elo) — higher is better. | 1656 | — |
Knowledge work GDPval-AA v2economically valuable knowledge work (v2, re-based Elo) | 1349 | — |
Chart reasoning CharXiv ReasoningCan the AI read and reason about complex charts and figures, not just text? Higher is better. | 84.2% | — |
Multimodal reasoning MMMU-ProA tougher version of MMMU — college-level questions that mix images, diagrams, and text together. Higher is better. | 83.6% | — |
Spatial reasoning Blueprint-Bench 2Can the AI reason about space and layout — for example, understanding a floor plan or blueprint? Higher is better. | 33.6% | — |
Long context MRCR v2 (8-needle) · 128k averageTests whether the AI can find specific details buried inside a very long document (around 128k tokens — roughly a long book). Higher is better. | 77.3% | — |
Long context MRCR v2 (8-needle) · 1M pointwiseTests whether the AI can find specific details buried inside an enormous document (around 1 million tokens — many books). Higher is better. | 26.6% | — |
Community preference Arena Elo (Text)Real people chat with two anonymous AIs side by side and vote for the answer they prefer. Votes become a chess-style Elo rating on arena.ai — it measures which AI people actually like, not test scores. Higher is better. | 1476 | — |
Community preference (code) Arena Elo (Code)Like the text arena, but people vote on which AI writes better code. The votes become a chess-style Elo rating on arena.ai. Higher is better. | 1499 | — |
| Overview | ||
| Company | SpaceXAI | |
| Release date | May 19 2026 | May 18 2026 |
| Access | Proprietary | Proprietary |
Which is better: Gemini 3.5 Flash or Composer 2.5?
Gemini 3.5 Flash and Composer 2.5 are evenly matched across the 4 benchmarks they both report (SWE-Bench Pro, CursorBench v3.2, CursorBench v3.1, Terminal-Bench 2.1). Composer 2.5 shipped 1 days before Gemini 3.5 Flash, so benchmark comparisons should account for the intervening progress.
Published specifications for these two models are limited — see each model page for the latest details.
On SWE-Bench Pro, Gemini 3.5 Flash leads at 55.1% vs Composer 2.5 at 54%. On CursorBench v3.2, Composer 2.5 leads at 56.1% vs Gemini 3.5 Flash at 48.8%. On CursorBench v3.1, Composer 2.5 leads at 63.2% vs Gemini 3.5 Flash at 49.8%. On Terminal-Bench 2.1, Gemini 3.5 Flash leads at 76.2% vs Composer 2.5 at 73%.
Frequently asked questions
Gemini 3.5 Flash was released by Google on May 19 2026.
Composer 2.5 was released by SpaceXAI on May 18 2026.
Gemini 3.5 Flash leads on SWE-Bench Pro — Gemini 3.5 Flash 55.1% vs Composer 2.5 54%.
Other comparisons
Gemini 3.5 FlashvsClaude Opus 5Composer 2.5vsClaude Opus 5Gemini 3.5 FlashvsGPT-5.6-CyberComposer 2.5vsGPT-5.6-CyberGemini 3.5 FlashvsMuse GlimmerComposer 2.5vsMuse GlimmerGemini 3.5 FlashvsDeepSeek-V4-Pro-0813Composer 2.5vsDeepSeek-V4-Pro-0813Gemini 3.5 FlashvsMistral Medium 3.5Composer 2.5vsMistral Medium 3.5Gemini 3.5 FlashvsKimi K3Composer 2.5vsKimi K3