Breaking

Anthropic Releases Claude Sonnet 5.5, Citing Epic Games Code Tests and a 70.6% Terminal-Bench Score

September 29, 2026 at 10:16 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioning the new model as a faster and cheaper workhorse for coding, agents, documents and design. The most visible gaming signal came from Epic Games, which tested the model on gameplay-system architecture reviews involving tens of thousands of lines of code.

Sep 28, 2026 · Anthropic

Sonnet moves closer to Opus

Claude Sonnet 5.5 targets everyday coding and agent work with faster runs, lower task cost and early game-development tests that reached large codebases.

Terminal-Bench 4.0 accuracy

70.6%

Sonnet 5.5

66.4%

Opus 5.5

10.3%

Sonnet 5

The coding benchmark is the clearest published jump: Sonnet 5.5 moves from the old Sonnet tier to near Opus-level results.

30%+

faster output than Sonnet 5

token context window

128K

maximum output tokens

$2/$10

input and output per 1M tokens

What changed for builders

Clear spec

best fit for bounded tasks

→

Fewer steps

less wasted tool use

→

Checked output

runtime and tests matter

Why teams like it

Near-Opus coding scores, faster iteration, fewer tool calls and credible early results in game, app and support workflows.

Where caution remains

Open-ended work still points to Opus, high effort can erode savings, and API migrations need checks.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.