At the moment, OpenAI rival Anthropic introduced Claude 4 fashions, that are considerably higher than Claude 3 in benchmarks, however we’re left dissatisfied with the identical 200,000 context window restrict.
In a weblog put up, Anthropic stated Claude Opus 4 is the corporate’s strongest mannequin, and it is also the most effective mannequin for coding within the business.
For instance, in SWE-bench (SWE is brief for Software program Engineering Benchmark), Claude Opus 4 scored 72.5 % and 43.2 on Terminal-bench.
“It delivers sustained efficiency on long-running duties that require targeted effort and 1000’s of steps, with the flexibility to work repeatedly for a number of hours, dramatically outperforming all Sonnet fashions and considerably increasing what AI brokers can accomplish,” Anthropic famous.
Whereas benchmarks put Claude 4 Sonnet and Opus forward of their predecessors and rivals like Gemini 2.5 Professional in coding, we’re nonetheless involved in regards to the mannequin’s 200,000 context window restrict.
This could possibly be one of many the explanation why Claude 4 fashions excel at coding and complex-solving duties in these benchmarks, as a result of these fashions are usually not being examined towards a big context.
For comparability, Google’s Gemini 2.5 Professional ships with a 1 million token context window and help for a 2 million context window can also be within the works.
ChatGPT’s 4.1 fashions additionally provide as much as a million context window.
| Mannequin | Description | Enter | Immediate Caching Write | Immediate Caching Learn | Output | Context Window | Batch Processing Low cost |
|---|---|---|---|---|---|---|---|
| Claude Opus 4 | Most clever mannequin for advanced duties | $15 / MTok | $18.75 / MTok | $1.50 / MTok | $75 / MTok | 200K | 50% low cost with batch processing |
| Claude Sonnet 4 | Optimum steadiness of intelligence, price, and pace | $3 / MTok | $3.75 / MTok | $0.30 / MTok | $15 / MTok | 200K | 50% low cost with batch processing |
Claude remains to be lagging behind the competitors in terms of the context window, which is essential in giant tasks.
Primarily based on an evaluation of 14M malicious actions, uncover the highest 10 MITRE ATT&CK strategies behind 93% of assaults and the right way to defend towards them.
