Benchmark updates
Recent capture activity across benchmarked coding agents. Each row shows the latest version hibench recorded, when it was captured, and how default token footprint changed versus the previous version.
Benchmark data last updated July 7, 2026 UTC
| Agent | Version | Captured | Tokens (o200k_base) | Δ tokens | Tools | Skills |
|---|---|---|---|---|---|---|
 OpenCode | v1.17.14 was v1.17.13 | July 7, 2026 | 6,671 | 0 +0% | 9 0 | 1 0 |
 Mistral Vibe | v2.19.0 was v2.18.4 | July 7, 2026 | 5,785 | -2,439 -30% | 9 0 | 1 0 |
 Kilo Code | v7.4.1 was v7.4.0 | July 7, 2026 | 9,426 | 0 +0% | 13 0 | 1 0 |
 Hermes Agent | v0.18.0 was v0.17.0 | July 7, 2026 | 12,223 | -919 -7% | 27 0 | 0 0 |
 Grok CLI | v0.2.89 was v0.2.88 | July 7, 2026 | 9,287 | 0 +0% | 24 0 | 7 0 |
 Copilot CLI | v1.0.68 was v1.0.67 | July 7, 2026 | 13,062 | -69 -1% | 15 0 | 1 0 |
 Droid | v0.164.1 was v0.164.0 | July 7, 2026 | 13,023 | +2 +0% | 10 0 | 20 0 |
 Devin | v3000.1.27 was v2026.8.18 | July 7, 2026 | 11,713 | +6 +0% | 23 0 | 2 0 |
 Cursor CLI | v2026.07.01-41b2de7 was v2026.06.29-2ad2186 | July 7, 2026 | 16,390 | 0 +0% | 15 0 | 0 0 |
 Cline | v3.0.38 was v3.0.37 | July 7, 2026 | 4,530 | +68 +2% | 25 0 | 0 0 |
 Claude Code | v2.1.202 was v2.1.201 | July 7, 2026 | 20,556 | +72 +0% | 25 0 | 14 0 |
 Pi | v0.80.3 was v0.80.2 | July 1, 2026 | 1,264 | +9 +1% | 4 0 | 0 0 |
 OpenClaw | v2026.6.11 was v2026.6.10 | July 1, 2026 | 18,863 | +118 +1% | 33 0 | 0 0 |
 Gemini CLI | v0.49.0 was v0.47.0 | June 27, 2026 | 8,513 | +48 +1% | 8 0 | 2 0 |
 Codex CLI | v0.142.0 was v0.141.0 | June 24, 2026 | 8,710 | +5 +0% | 11 0 | 5 0 |
 OpenHands | v1.16.0 was v1.15.0 | June 19, 2026 | 11,868 | -1,000 -8% | 7 +1 | 49 0 |
Sorted by most recent capture date. Token deltas compare the latest version to the immediately prior captured version using the fixed o200k_base tokenizer.