| Developer |
Anthropic |
Google DeepMind |
| Released |
2026-02-17 |
2026-02-19 |
| Input Price |
$3/1M tokens |
$2/1M tokens |
| Output Price |
$15/1M tokens |
$12/1M tokens |
| Context Window |
1M tokens (beta) |
1M tokens (standard) |
| Output Tokens |
128K |
Not disclosed |
| Thinking Mode |
Adaptive thinking |
Three-tier thinking system — Low, Medium, High |
| Multimodal |
Yes — text and images |
Yes — text, images, audio, video |
| API Available |
Yes — API, Bedrock, Vertex AI, Foundry |
Yes — AI Studio, Gemini API, Vertex AI |
| Best For |
Production coding, large codebase understanding, long-form writing, instruction-following workflows, enterprise RAG, GitHub Copilot agent base model |
Abstract reasoning, scientific research, multimodal workflows with audio and video, high-volume API pipelines, Google ecosystem integration |
| SWE-Bench Verified |
79.6% |
80.6% ✓ |
| SWE-Bench Pro |
68.1% |
79.6% ✓ |
| ARC-AGI-2 |
58.3% |
77.1% ✓ |
| GPQA Diamond |
74.1% |
94.3% ✓ |
| GDPval-AA (Knowledge Work) |
Leading ✓ |
316 Elo behind |
| Terminal-Bench 2.0 |
65.4% |
74.8% ✓ |
| Long-Context Coherence |
Strong ✓ |
Standard |
| Human Writing Preference |
Preferred ✓ |
Formulaic |
By Joe ~ July 2, 2026