#terminal-bench

Ricursive released Ripple-65B Preview on November 4, 2026: a 65B dense model, 1M context, Apache 2.0. What the vendor claims, what is verifiable, what it costs.

Terminal-Bench 4.0 changes the test behind agent scores. Decide when results need rerunning, regrading or reuse before comparing coding agents for your team.

OSWorld measures desktop GUI control, Terminal-Bench measures CLI work. Same-named benchmarks differ by version and harness, so check before you compare.

Google markets Gemini 3.5 Flash-Lite for subagents at $0.30/$2.50 per Mtok and 350 tok/s — a 10-worker fleet can cost a fraction of a flagship fleet.

poolside's 118B open-weight Laguna S 2.1 runs on one DGX Spark and, on its own harness, tops rivals 10x its size on Terminal-Bench — trained in under 9 weeks.

Opus 4.8 tops the Artificial Analysis index, but GPT-5.5 still leads Terminal-Bench. An evidence-graded roundup of the first 48 hours of independent evals.

Agentic coding head-to-head: Gemini 3.5 Flash vs GPT-5.5 vs Opus 4.7. MCP Atlas, SWE-Bench Pro, Terminal-Bench, plus Antigravity 2.0 launch context.

OpenAI's GPT-5.5 ships April 23, 2026 with 1M context, Thinking and Pro variants, 82.7% Terminal-Bench, and same latency as GPT-5.4. Pricing inside.

Alibaba's Qwen3.6-Max-Preview tops six coding benchmarks including SWE-bench Pro and Terminal-Bench 2.0. Closed-weights pivot and agency playbook inside.

Cursor's Composer 1.5 scales reinforcement learning 20x to score 47.9% on Terminal-Bench 2.0 with adaptive thinking and self-summarization. Full analysis.