Compare AI models
| Specifications | ||
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 400k | — |
| API pricingUSD per 1M tokens · lower wins | ||
Input priceWhat you pay for everything you send the model — your question, plus any documents or earlier conversation you include with it. | $0.75 | — |
Output priceWhat you pay for the text the model writes back. It is normally the dearer half: producing an answer costs more than reading one. | $4.50 | — |
Cached input priceA reduced rate for text you send over and over. If every request starts with the same instructions or the same document, the provider keeps a copy ready and charges less to read it again. | $0.075 | — |
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | $0.375OpenAI | — |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | $2.25OpenAI | — |
| $375OpenAI list price | — | |
| Benchmarks | ||
SWE-Bench ProAgentic coding — Can 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. | 54.4% | 54% |
Terminal-Bench 2.0Agentic terminal coding — Can 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. | 60% | 69.3% |
BullshitBench v2Nonsense detection — Given 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. | 32% | — |
ProgramBenchProgram reconstruction — The AI receives a working program and its documentation, then builds a replacement from scratch without the original source code, internet access or decompilation. The score is the percentage of 200 programs that pass every behavioral test. We record each model's best published mini-SWE-agent result, including higher reasoning efforts where available. Partial test-pass rates and almost-solved programs do not count toward this score. Equal scores share a rank here; the official board also uses partial progress to break ties. Higher is better. | 0% | — |
SWE-Bench MultilingualMultilingual coding — Like 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% |
CursorBench 4.0Agentic coding — Cursor's own test of coding agents on ambiguous, multi-file tasks taken from real Cursor sessions — editing, refactoring, investigating a codebase, understanding what the user meant, managing jobs and following a design. Cursor runs each model at several reasoning efforts; each release here carries the score of its best listed effort. Scores aren't comparable with earlier CursorBench versions. Higher is better. | — | 27.7% |
DeepSWE 1.0Agentic coding — Artificial 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% |
Next.js EvalsNext.js coding — Vercel'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. | — | 82% |
Supabase Evals · with skillsSupabase coding — Supabase's own open benchmark: a coding agent is dropped into a real Supabase project and asked to do real work — set up a schema, fix a broken security policy, debug an Edge Function — and every run is checked against a live Supabase stack. This is the headline number, where the agent has Supabase's own skills loaded, as most people building on Supabase would. The score is the share of scenarios it got right. Higher is better. | 65.2% | — |
Supabase Evals · no skillsSupabase coding — The same Supabase scenarios, but with none of Supabase's skills loaded — so it measures what the model already knows about building on Supabase, rather than how well it follows Supabase's supplied instructions. Higher is better. | 55.1% | — |
Terminal-Bench 2.1Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better. | — | 73% |
MCP Atlas · original harnessMulti-step tool use — pass rate under the original tool-use harness, before Scale changed the judge and retry handling in April 2026 | 57.7% | — |
ToolathlonGeneral tool use — Tests how well the AI uses everyday real-world tools and apps to get things done. Higher is better. | 42.9% | — |
BU BenchBrowser agent — Can 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. | 36% | — |
Humanity's Last Exam · no toolsMultidisciplinary reasoning — Humanity'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. | 28.2% | — |
Humanity's Last Exam · with toolsMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects. “With tools” means the AI is allowed to search the web or run code while answering. Higher is better. | 41.5% | — |
ARC-AGI-2Abstract reasoning — Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better. | 18.9% | — |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 88% | — |
OSWorld-VerifiedAgentic computer use — Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better. | 72.1% | — |
MMMU-Pro · with toolsMultimodal reasoning — multimodal understanding and reasoning with access to tools | 78% | — |
MMMU-ProMultimodal reasoning — A tougher version of MMMU — college-level questions that mix images, diagrams, and text together. Higher is better. | 76.6% | — |
GraphWalks BFS (<128K)Long context — breadth-first traversal accuracy for inputs shorter than 128K tokens | 76.3% | — |
GraphWalks parents (<128K)Long context — parent-node retrieval accuracy for inputs shorter than 128K tokens | 71.5% | — |
threejsevalCommunity preference (Three.js) — Every model gets the same prompt — "the Eiffel Tower", "a glass fishbowl", "a robot arm picking toys into a box" — and builds a 3D scene in Three.js. Real people then see two scenes side by side, names hidden, and vote for the one they prefer. The votes become a chess-style Elo rating on threejseval.com, averaged across all the prompts. It measures whether the scene looks and moves right to a human eye, not whether the code passes a test. Higher is better. | — | 1083 |
| Overview | ||
| Company | OpenAI | SpaceXSI |
| Release date | Mar 17 2026 | May 18 2026 |
| Access | Closed | Closed |
| Model details | View model | View model |
Other comparisons
GPT-5.4 minivsClaude Fable 5.2Composer 2.5vsClaude Fable 5.2GPT-5.4 minivsGemini 4 CarbonComposer 2.5vsGemini 4 CarbonGPT-5.4 minivsMuse EmberComposer 2.5vsMuse EmberGPT-5.4 minivsDeepSeek-V4.1-Flash-1017Composer 2.5vsDeepSeek-V4.1-Flash-1017GPT-5.4 minivsMistral Large 4Composer 2.5vsMistral Large 4GPT-5.4 minivsKimi K3.1 ProComposer 2.5vsKimi K3.1 ProFrequently asked questions
GPT-5.4 mini and Composer 2.5 are evenly matched across the 2 benchmarks they both report (SWE-Bench Pro, Terminal-Bench 2.0). Only GPT-5.4 mini has a verified first-party price for an available API: $0.75 per million input tokens and $4.50 per million output tokens. No currently available pay-as-you-go API rate is tracked for Composer 2.5. GPT-5.4 mini shipped 62 days before Composer 2.5, 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, GPT-5.4 mini leads at 54.4% vs Composer 2.5 at 54%. On Terminal-Bench 2.0, Composer 2.5 leads at 69.3% vs GPT-5.4 mini at 60%.