Average 40% cheaper per run across 20 tasks

20 tasks, three runs each. Each coding agent is measured against its own baseline run without SourceIndex. Codex(Preview) is reported on its own: SourceIndex support for Codex is not fully ready yet, so it ran a separate set of 22 tasks from the same repositories.

Average per run

Cost and minutes per run, averaged over the tasks both setups solved in every run. Claude Code and OpenCode ran the same 20 tasks.

Tool · model Resolved Cost / run Cost saving Minutes / run Time saving
Claude Code · Opus 4.660/60→59/60$1.258→$0.77438%4.56→3.3626%
OpenCode · GPT-5.556/60→56/60$1.327→$0.77342%5.08→4.5311%

Codex: a different set of 22 tasks, and a preview. Compare it with its own baseline only.

Tool · model Resolved Cost / run Cost saving Minutes / run Time saving
Codex(Preview) · GPT-5.557/66→57/66$0.977→$0.63935%5.21→4.3716%

Averages cover the tasks solved in every run by both setups: 19 of 20 for Claude Code, 17 of 20 for OpenCode, and 19 of 22 for Codex(Preview).

Claude Code

Cost Saving 38% & Time Saving 26%

Claude Code 2.1.283 driving Opus 4.6 at high reasoning effort. Per task, mean of three runs.

Cost per run Claude Code Claude Code + SourceIndex
$0 $1 $2 $3 NodeBB-1 — Claude Code $0.83 → + SourceIndex $0.64 (22% less) −22% NodeBB-1 NodeBB-2 — Claude Code $0.38 → + SourceIndex $0.33 (15% less) −15% NodeBB-2 NodeBB-3 — Claude Code $3.21 → + SourceIndex $0.91 (72% less) −72% NodeBB-3 ansible-1 — Claude Code $2.22 → + SourceIndex $1.41 (37% less) −37% ansible-1 element-web-1 — Claude Code $1.41 → + SourceIndex $1.17 (17% less) −17% element-web-1 element-web-2 — Claude Code $1.42 → + SourceIndex $0.86 (40% less) −40% element-web-2 flipt-1 — Claude Code $0.94 → + SourceIndex $0.47 (50% less) −50% flipt-1 flipt-2 — Claude Code $0.75 → + SourceIndex $0.50 (33% less) −33% flipt-2 flipt-3 — Claude Code $0.96 → + SourceIndex $0.60 (38% less) −38% flipt-3 flipt-4 — Claude Code $1.18 → + SourceIndex $0.61 (48% less) −48% flipt-4 navidrome-1 — Claude Code $1.40 → + SourceIndex $1.00 (29% less) −29% navidrome-1 navidrome-2 — Claude Code $1.60 → + SourceIndex $0.77 (52% less) −52% navidrome-2 qutebrowser-1 — Claude Code $0.73 → + SourceIndex $0.49 (33% less) −33% qutebrowser-1 qutebrowser-2 — Claude Code $0.49 → + SourceIndex $0.58 (19% more) +19% qutebrowser-2 teleport-1 — Claude Code $0.88 → + SourceIndex $0.64 (27% less) −27% teleport-1 tutanota-1 — Claude Code $1.15 → + SourceIndex $0.88 (23% less) −23% tutanota-1 vuls-1 — Claude Code $1.72 → + SourceIndex $1.28 (26% less) −26% vuls-1 vuls-2 — Claude Code $0.86 → + SourceIndex $0.53 (38% less) −38% vuls-2 webclients-1 — Claude Code $1.15 → + SourceIndex $0.84 (27% less) −27% webclients-1 webclients-2 — Claude Code $1.76 → + SourceIndex $1.08 (39% less) −39% webclients-2

Row under the columns: SourceIndex vs baseline cost (− less, + more). Hover a task for exact values.

Minutes per run Claude Code Claude Code + SourceIndex
0 2 4 6 8 NodeBB-1 — Claude Code 2.9 min → + SourceIndex 2.5 min (13% less) −13% NodeBB-1 NodeBB-2 — Claude Code 1.2 min → + SourceIndex 1.2 min (same) 0% NodeBB-2 NodeBB-3 — Claude Code 6.6 min → + SourceIndex 3.3 min (51% less) −51% NodeBB-3 ansible-1 — Claude Code 5.6 min → + SourceIndex 5.7 min (2% more) +2% ansible-1 element-web-1 — Claude Code 6.8 min → + SourceIndex 5.1 min (26% less) −26% element-web-1 element-web-2 — Claude Code 6.1 min → + SourceIndex 3.9 min (36% less) −36% element-web-2 flipt-1 — Claude Code 3.5 min → + SourceIndex 2.0 min (42% less) −42% flipt-1 flipt-2 — Claude Code 3.9 min → + SourceIndex 2.6 min (33% less) −33% flipt-2 flipt-3 — Claude Code 3.7 min → + SourceIndex 2.0 min (45% less) −45% flipt-3 flipt-4 — Claude Code 5.2 min → + SourceIndex 3.2 min (38% less) −38% flipt-4 navidrome-1 — Claude Code 5.5 min → + SourceIndex 3.9 min (29% less) −29% navidrome-1 navidrome-2 — Claude Code 6.1 min → + SourceIndex 3.4 min (45% less) −45% navidrome-2 qutebrowser-1 — Claude Code 2.8 min → + SourceIndex 2.2 min (22% less) −22% qutebrowser-1 qutebrowser-2 — Claude Code 1.7 min → + SourceIndex 2.0 min (13% more) +13% qutebrowser-2 teleport-1 — Claude Code 3.3 min → + SourceIndex 2.5 min (23% less) −23% teleport-1 tutanota-1 — Claude Code 4.0 min → + SourceIndex 3.1 min (22% less) −22% tutanota-1 vuls-1 — Claude Code 6.4 min → + SourceIndex 6.4 min (1% more) +1% vuls-1 vuls-2 — Claude Code 3.1 min → + SourceIndex 2.6 min (14% less) −14% vuls-2 webclients-1 — Claude Code 5.1 min → + SourceIndex 4.3 min (14% less) −14% webclients-1 webclients-2 — Claude Code 7.2 min → + SourceIndex 5.0 min (31% less) −31% webclients-2

Row under the columns: SourceIndex vs baseline time (− less, + more). Hover a task for exact values.

OpenCode

Cost Saving 42% & Time Saving 11%

OpenCode 2.0.22 driving GPT-5.5 at high reasoning effort. Per task, mean of three runs.

Cost per run OpenCode OpenCode + SourceIndex
$0 $1 $2 $3 NodeBB-1 — OpenCode $0.99 → + SourceIndex $0.45 (55% less) −55% NodeBB-1 NodeBB-2 — OpenCode $0.38 → + SourceIndex $0.28 (25% less) −25% NodeBB-2 NodeBB-3 — OpenCode $3.15 → + SourceIndex $1.75 (45% less) −45% NodeBB-3 ansible-1 — OpenCode $2.14 → + SourceIndex $1.36 (36% less) −36% ansible-1 element-web-1 — OpenCode $1.61 → + SourceIndex $0.82 (49% less) −49% element-web-1 element-web-2 — OpenCode $2.49 → + SourceIndex $1.28 (48% less) −48% element-web-2 flipt-1 — OpenCode $1.67 → + SourceIndex $0.49 (71% less) −71% flipt-1 flipt-2 — OpenCode $0.67 → + SourceIndex $0.57 (14% less) −14% flipt-2 flipt-3 — OpenCode $1.95 → + SourceIndex $0.75 (62% less) −62% flipt-3 flipt-4 — OpenCode $1.74 → + SourceIndex $0.62 (65% less) −65% flipt-4 navidrome-1 — OpenCode $1.95 → + SourceIndex $1.41 (28% less) −28% navidrome-1 navidrome-2 — OpenCode $1.04 → + SourceIndex $0.75 (28% less) −28% navidrome-2 qutebrowser-1 — OpenCode $1.03 → + SourceIndex $0.47 (55% less) −55% qutebrowser-1 qutebrowser-2 — OpenCode $0.35 → + SourceIndex $0.28 (20% less) −20% qutebrowser-2 teleport-1 — OpenCode $0.43 → + SourceIndex $0.55 (28% more) +28% teleport-1 tutanota-1 — OpenCode $1.16 → + SourceIndex $0.68 (41% less) −41% tutanota-1 vuls-1 — OpenCode $1.78 → + SourceIndex $0.91 (49% less) −49% vuls-1 vuls-2 — OpenCode $0.53 → + SourceIndex $0.50 (5% less) −5% vuls-2 webclients-1 — OpenCode $1.26 → + SourceIndex $0.83 (34% less) −34% webclients-1 webclients-2 — OpenCode $2.34 → + SourceIndex $1.75 (25% less) −25% webclients-2

Row under the columns: SourceIndex vs baseline cost (− less, + more). Hover a task for exact values.

Minutes per run OpenCode OpenCode + SourceIndex
0 2 4 6 8 NodeBB-1 — OpenCode 3.5 min → + SourceIndex 3.2 min (11% less) −11% NodeBB-1 NodeBB-2 — OpenCode 2.1 min → + SourceIndex 1.8 min (12% less) −12% NodeBB-2 NodeBB-3 — OpenCode 7.0 min → + SourceIndex 5.6 min (20% less) −20% NodeBB-3 ansible-1 — OpenCode 6.4 min → + SourceIndex 6.9 min (8% more) +8% ansible-1 element-web-1 — OpenCode 6.9 min → + SourceIndex 5.1 min (25% less) −25% element-web-1 element-web-2 — OpenCode 7.8 min → + SourceIndex 6.5 min (16% less) −16% element-web-2 flipt-1 — OpenCode 4.6 min → + SourceIndex 3.6 min (22% less) −22% flipt-1 flipt-2 — OpenCode 5.0 min → + SourceIndex 3.9 min (22% less) −22% flipt-2 flipt-3 — OpenCode 7.3 min → + SourceIndex 4.4 min (40% less) −40% flipt-3 flipt-4 — OpenCode 6.0 min → + SourceIndex 3.7 min (38% less) −38% flipt-4 navidrome-1 — OpenCode 6.8 min → + SourceIndex 6.5 min (4% less) −4% navidrome-1 navidrome-2 — OpenCode 4.5 min → + SourceIndex 4.1 min (7% less) −7% navidrome-2 qutebrowser-1 — OpenCode 4.8 min → + SourceIndex 3.9 min (19% less) −19% qutebrowser-1 qutebrowser-2 — OpenCode 2.4 min → + SourceIndex 2.5 min (5% more) +5% qutebrowser-2 teleport-1 — OpenCode 2.8 min → + SourceIndex 4.1 min (44% more) +44% teleport-1 tutanota-1 — OpenCode 4.4 min → + SourceIndex 3.9 min (11% less) −11% tutanota-1 vuls-1 — OpenCode 7.0 min → + SourceIndex 4.5 min (36% less) −36% vuls-1 vuls-2 — OpenCode 2.8 min → + SourceIndex 3.7 min (35% more) +35% vuls-2 webclients-1 — OpenCode 5.2 min → + SourceIndex 4.8 min (7% less) −7% webclients-1 webclients-2 — OpenCode 7.7 min → + SourceIndex 8.2 min (7% more) +7% webclients-2

Row under the columns: SourceIndex vs baseline time (− less, + more). Hover a task for exact values.

Codex(Preview)

Cost Saving 35% & Time Saving 16%

Preview only. Codex 0.156.1 driving GPT-5.5 at high reasoning effort. Per task, mean of three runs.

Not fully ready yet. SourceIndex support for Codex is still in preview. Codex calls SourceIndex as an MCP tool, not through the SourceIndex CLI.

Different tasks. Codex ran a separate set of 22 tasks from the same repositories. Task labels are numbered within each set.

Cost per run Codex Codex + SourceIndex
$0 $0.5 $1 $1.5 $2 NodeBB-1 — Codex $1.60 → + SourceIndex $1.07 (33% less) −33% NodeBB-1 NodeBB-2 — Codex $1.48 → + SourceIndex $0.76 (49% less) −49% NodeBB-2 NodeBB-3 — Codex $1.49 → + SourceIndex $0.69 (53% less) −53% NodeBB-3 ansible-1 — Codex $0.41 → + SourceIndex $0.29 (28% less) −28% ansible-1 ansible-2 — Codex $1.21 → + SourceIndex $0.84 (31% less) −31% ansible-2 ansible-3 — Codex $0.86 → + SourceIndex $0.65 (24% less) −24% ansible-3 element-web-1 — Codex $0.76 → + SourceIndex $0.36 (52% less) −52% element-web-1 element-web-2 — Codex $0.46 → + SourceIndex $0.31 (33% less) −33% element-web-2 flipt-1 — Codex $1.51 → + SourceIndex $0.63 (58% less) −58% flipt-1 flipt-2 — Codex $0.47 → + SourceIndex $0.47 (same) 0% flipt-2 navidrome-1 — Codex $0.95 → + SourceIndex $0.24 (74% less) −74% navidrome-1 navidrome-2 — Codex $0.43 → + SourceIndex $0.42 (3% less) −3% navidrome-2 qutebrowser-1 — Codex $0.89 → + SourceIndex $0.46 (48% less) −48% qutebrowser-1 qutebrowser-2 — Codex $1.43 → + SourceIndex $1.25 (13% less) −13% qutebrowser-2 teleport-1 — Codex $1.49 → + SourceIndex $0.91 (39% less) −39% teleport-1 tutanota-1 — Codex $0.82 → + SourceIndex $0.54 (34% less) −34% tutanota-1 vuls-1 — Codex $0.82 → + SourceIndex $0.41 (50% less) −50% vuls-1 vuls-2 — Codex $1.41 → + SourceIndex $1.30 (7% less) −7% vuls-2 webclients-1 — Codex $0.77 → + SourceIndex $0.50 (35% less) −35% webclients-1 webclients-2 — Codex $1.72 → + SourceIndex $1.77 (3% more) +3% webclients-2 webclients-3 — Codex $0.39 → + SourceIndex $0.17 (58% less) −58% webclients-3 webclients-4 — Codex $0.45 → + SourceIndex $0.28 (38% less) −38% webclients-4

Row under the columns: SourceIndex vs baseline cost (− less, + more). Hover a task for exact values.

Minutes per run Codex Codex + SourceIndex
0 2 4 6 8 10 NodeBB-1 — Codex 7.9 min → + SourceIndex 7.1 min (10% less) −10% NodeBB-1 NodeBB-2 — Codex 7.4 min → + SourceIndex 5.5 min (26% less) −26% NodeBB-2 NodeBB-3 — Codex 7.6 min → + SourceIndex 5.3 min (31% less) −31% NodeBB-3 ansible-1 — Codex 2.9 min → + SourceIndex 2.4 min (17% less) −17% ansible-1 ansible-2 — Codex 7.4 min → + SourceIndex 5.9 min (21% less) −21% ansible-2 ansible-3 — Codex 5.2 min → + SourceIndex 4.7 min (10% less) −10% ansible-3 element-web-1 — Codex 4.9 min → + SourceIndex 3.0 min (39% less) −39% element-web-1 element-web-2 — Codex 3.1 min → + SourceIndex 2.9 min (8% less) −8% element-web-2 flipt-1 — Codex 6.8 min → + SourceIndex 4.2 min (39% less) −39% flipt-1 flipt-2 — Codex 2.5 min → + SourceIndex 3.1 min (24% more) +24% flipt-2 navidrome-1 — Codex 4.4 min → + SourceIndex 2.0 min (55% less) −55% navidrome-1 navidrome-2 — Codex 2.4 min → + SourceIndex 3.3 min (37% more) +37% navidrome-2 qutebrowser-1 — Codex 5.2 min → + SourceIndex 3.4 min (35% less) −35% qutebrowser-1 qutebrowser-2 — Codex 6.0 min → + SourceIndex 6.9 min (14% more) +14% qutebrowser-2 teleport-1 — Codex 9.7 min → + SourceIndex 6.9 min (28% less) −28% teleport-1 tutanota-1 — Codex 4.6 min → + SourceIndex 3.8 min (17% less) −17% tutanota-1 vuls-1 — Codex 5.8 min → + SourceIndex 3.3 min (42% less) −42% vuls-1 vuls-2 — Codex 6.2 min → + SourceIndex 7.2 min (17% more) +17% vuls-2 webclients-1 — Codex 4.4 min → + SourceIndex 3.8 min (15% less) −15% webclients-1 webclients-2 — Codex 8.5 min → + SourceIndex 10.0 min (19% more) +19% webclients-2 webclients-3 — Codex 1.5 min → + SourceIndex 1.3 min (9% less) −9% webclients-3 webclients-4 — Codex 2.7 min → + SourceIndex 2.1 min (20% less) −20% webclients-4

Row under the columns: SourceIndex vs baseline time (− less, + more). Hover a task for exact values.