Compare: LLMs & GPUs
Compare up to 6 LLMs on specs and benchmark scores, or put two GPUs side by side on VRAM, bandwidth, power and cost.
Specifications
| Spec | ||||
|---|---|---|---|---|
| Vendor | Anthropic | OpenAI | Google DeepMind | DeepSeek |
| License | Proprietary | Proprietary | Proprietary | Open · MIT |
| Release | 2026-11-09 | 2026-11-17 | 2026-11-10 | 2026-08-13 |
| Parameters | Undisclosed | Undisclosed | Undisclosed | MoE 1.6T / A49B |
| Context | 1M | 1.05M | 1.05M | 1.02M |
| Max Output | 128K | 128K | 66K | 384K |
| Modalities | Text + Image + PDF | Text + Image + PDF | Text + Image + Video + Audio + PDF | Text |
| Reasoning | ✅ | ✅ | ✅ | ✅ |
| Input /1M | $8 | $8 | $2 | $1.32 |
| Output /1M | $40 | $40 | $10 | $3.96 |
| Cached /1M | $0.2 | $0.4 | $0.1 | $0.044 |
8-Axis Capability Radar
Claude Fable 5.2(wins 8)GPT-6.2 Astra(wins 2)Gemini 4 Carbon(wins 2)DeepSeek-V4-Pro (1.6T MoE)(wins 0)
| Dimension | ||||
|---|---|---|---|---|
| 🏆 Overall | 🏆 63.7 #1 | 62.4 #3 | 60.7 #5 | 45.1 #64 |
| 🧠 Reasoning | 🏆 63.7 #1 | 61.0 #3 | 58.8 #5 | 45.5 #59 |
| 💻 Coding | 🏆 55.0 #1 | 52.5 #2 | 50.6 #5 | 35.2 #58 |
| 📐 Math | 🏆 58.7 #1 | — | 25.0 #107 | 45.2 #25 |
| 🤖 Agents | 🏆 50.6 #2 | 49.9 #3 | 47.9 #4 | 27.4 #71 |
| 🛠️ Tool Calling | 44.7 #3 | 🏆 46.2 #1 | 41.4 #5 | 29.3 #58 |
| 🎨 Multimodal | 🏆 49.5 #1 | — | 42.8 #9 | — |
| 👁️ Vision | 🏆 53.6 #1 | — | 43.1 #12 | 35.4 #44 |
| 📜 Long Context | — | — | 🏆 42.3 #2 | 34.7 #13 |
| 🖌️ Frontend Dev | — | — | 31.5 #30 | 🏆 40.2 #8 |
| 🌐 Multilingual | 38.7 #20 | — | — | 🏆 45.1 #5 |
| 🔎 Search | — | — | — | 🏆 31.2 #23 |
| ⚛️ Physics | — | 🏆 53.5 #1 | — | 42.5 #30 |
| 🩺 Healthcare | 🏆 44.0 #8 | 40.0 #21 | — | 43.7 #9 |
| 💹 Finance | — | — | — | 🏆 43.7 #7 |
| ⚖️ Legal | — | — | — | 🏆 43.7 #7 |
📋 Benchmark-by-benchmark (12)
Darker = closer to best; — = not reported
| Benchmark | Claude Fable 5.2 | GPT-6.2 Astra | Gemini 4 Carbon | DeepSeek-V4-Pro (1.6T MoE) |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 Terminal-Bench-Science 0.1 is a Stanford-led community benchmark of 70 tasks drawn from scientific research workflows across the life, physical, earth, mathematical, and engineering sciences. | 74.6% | 🏆 77.1% | 64.8% | — |
| GPQA A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy. | — | 🏆 97% | — | 90.1% |
| Humanity's Last Exam Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions. | 🏆 73.2% | — | — | 48.2% |
| SWE-Bench Pro SWE-Bench Pro is an advanced version of SWE-Bench that evaluates language models on complex, real-world software engineering tasks requiring extended reasoning and multi-step problem solving. | 🏆 93% | — | — | 55.4% |
| SWE-bench Multilingual A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++. | 🏆 96.6% | — | — | 76.2% |
| AutomationBench 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. | 51.4% | — | 🏆 53.9% | — |
| HealthBench Professional HealthBench Professional evaluates model capability and safety for clinician use cases using real clinician-style chats and physician-authored grading rubrics. | 🏆 67.7% | 66.3% | — | — |
| SEC-bench Pro SEC-bench Pro is a self-evolving software-security benchmark that measures agent bug hunting on critical, high-complexity systems. | — | 86.8% | 🏆 88.2% | — |
| Chartography (no tools) Chartography visual chart recognition evaluated without tools, distinct from tool-assisted results. | 🏆 73.4% | — | 72.5% | — |
| OSWorld 2.1 (partial) Partial-credit computer-use score on OSWorld 2.1, as reported in the Claude Sonnet 5.5 release. Separate from OSWorld 2.0 and strict completion scores. | 🏆 89.3% | — | 81.5% | — |
| FrontierSWE V2 FrontierSWE V2 evaluates agents on open-ended technical projects and reports mean@5. Kept separate from FrontierSWE V1 dominance-score results, which are not comparable. | 🏆 69.9% | — | 61.2% | — |
| FrontierCode 2.0 FrontierCode 2.0 is Cognition's private benchmark assessing end-to-end code quality of coding-agent changes, including whether a repository maintainer would merge the agent's change. | 🏆 49.1% | 44.2% | — | — |