Claude Sonnet 5.5 Is Now Live on VEGA
Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family after Opus 5.5. It is a direct upgrade over Sonnet 5 for ordinary daily work: fixing bugs, writing clean documents, slides, and spreadsheets, and reading through code.
It is already live on our platform. Claude Sonnet 5.5 is available on vega.chat in the model picker. It is also available on api.vega.chat, our API for vibe coding and agents, at the best price.
What changed since Sonnet 5
The coding numbers show the clearest jump. On Terminal-Bench 4.0, an agentic coding evaluation, the model scores 70.6% against 10.3% for Sonnet 5. On CursorBench, built from real coding sessions, it reaches 55.5% against 34.1%. Early testers report that Sonnet 5.5 reads an unfamiliar codebase faster and batches tool calls together, so tasks finish in fewer steps.
Knowledge work moved up as well. On GDPval-AA, a test of real tasks across dozens of occupations, Sonnet 5.5 scores 1844 against 1449 for Sonnet 5, close to Opus 5.5 at 1846. Anthropic also points to its eye for design: interfaces come out more polished, and slide decks from a template need little editing.
Output streams more than 30% faster than Sonnet 5. One detail worth mentioning: it is the first Sonnet to finish Pokemon Red working only from screenshots.
Benchmarks
| Benchmark | Sonnet 5.5 | Sonnet 5 | Opus 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 70.6% | 10.3% | 66.4% |
| FrontierCode 1.1 (Main) | 46.2% | 42.4% | 54.4% |
| CursorBench 4.0 | 55.5% | 34.1% | 57.8% |
| GDPval-AA v2.1 (knowledge work) | 1844 | 1449 | 1846 |
| Humanity's Last Exam (with tools) | 64.5% | 54.9% | 67.7% |
| OSWorld 2.1 (computer use) | 80.1% | 57.0% | 81.8% |
| Chartography (chart recognition) | 61.6% | 15.6% | 64.4% |
Anthropic notes that benchmarks tell part of the story. In its own testing, Opus 5.5 stays stronger at complex open-ended work that needs sustained judgment. Sonnet 5.5 fits best at lower effort settings, where it answers quickly and keeps quality high.
Safety
On the automated behavioral audit, Sonnet 5.5 matches or improves on Sonnet 5 on most measures. It is the first Sonnet to launch with cyber safeguards of the class built for the most capable models, since its cybersecurity capabilities are comparable to Opus 5. Biology safeguards stay the same as Sonnet 5. Both target a narrow set of high-risk requests, while routine development and most life sciences work are unaffected.
Try it on VEGA
Opus 5.5 remains the pick for the hardest open-ended projects. Sonnet 5.5 sits next to it for fast iteration, well-scoped tasks, code, and documents you use every day. Haiku 5.5 will complete the family later for high-volume work.
Open vega.chat, pick Claude Sonnet 5.5 in the model list, and send your first task. If you build with agents or code assistants, connect through api.vega.chat.