Loading the page.
./stacks · catalog / pairings
snapshot: data 23.09.2026
Stacks and pairings
An agent+model pairing - not a model on its own: Claude Code with Opus 4.8 has different cost, speed, and quality than the same Claude Code with Fable 5. Below - 26 pairings with task cost and time from RuBench, AA Coding Agent Index, Terminal-Bench 2.1, and DeepSWE measurements.
- 26
- pairings
- agent + model, current config
- 4
- with RuBench
- our run, open trajectories
- 62.2
- best AA Index
- Claude Code · Fable 5.1 · max*
full pairing rankingfull table, all columns, sortingRuBenchmethodology and run on Russian technical specs
catalog · 26 pairings
Each card is a pairing component: click takes you to the tool or model
Claude Code
Fable 5.1 · max*
62.2
AA Index
$12.39
Price/task
34.831668756875715 min
Time/task
62.2
AA Index
57.9
T-Bench
Codex
GPT-6 Astra · max
61.6
AA Index
$7.47
Price/task
29.36721710671066 min
Time/task
61.6
AA Index
58.2
T-Bench
73%
DeepSWE
Claude Code
Opus 5 · max
59.7
AA Index
$10.79
Price/task
41.92567059955995 min
Time/task
59.7
AA Index
74%
DeepSWE
Codex
GPT-6 Sol · max
56.7
AA Index
$2.99
Price/task
22.300878419508642 min
Time/task
Grok Build
Grok 4.7 · xhigh
56.3
AA Index
$8.82
Price/task
39.239461441144094 min
Time/task
56.3
AA Index
37.6
T-Bench
OpenCode
GLM-5.3
53.6
AA Index
$4.24
Price/task
48.13259420608729 min
Time/task
Claude Code
Opus 4.8 · max
78.7%
RuBench · our run
78.7%
RuBench
23.6
T-Bench
59%
DeepSWE
Codex
GPT-5.5 · xhigh
66.7%
RuBench · our run
66.7%
RuBench
67%
DeepSWE
Claude Code
Sonnet 5 · xhigh
74.7%
RuBench · our run
Claude Code
Haiku 4.5 · xhigh
53.3%
RuBench · our run
Claude Code
Opus 5 · xhigh
51.8
T-Bench
51.8
T-Bench
74%
DeepSWE
Claude Code
Fable 5 · max*
44.6
T-Bench
44.6
T-Bench
70%
DeepSWE
Claude Code
GLM-5.3
41.8
T-Bench
41.8
T-Bench
69%
DeepSWE
Codex
GPT-5.6 Sol · max
37.3
T-Bench
37.3
T-Bench
73%
DeepSWE
Codex
GPT-5.6 Terra · max
21.5
T-Bench
21.5
T-Bench
70%
DeepSWE
Grok Build
Grok 4.6 · high
20.3
T-Bench
20.3
T-Bench
65%
DeepSWE
Codex
GPT-5.6 Luna · max
17.3
T-Bench
17.3
T-Bench
67%
DeepSWE
Claude Code
Sonnet 5 · max
12.4
T-Bench
12.4
T-Bench
54%
DeepSWE
Grok Build
Grok 4.5 · high
12.4
T-Bench
12.4
T-Bench
54%
DeepSWE
Kimi Code CLI
Kimi K3
69%
DeepSWE
OpenCode
Gemini 3.8 Flash · high
74%
DeepSWE
Grok Build
Grok 4.6 · xhigh
67%
DeepSWE
Claude Code
DeepSeek V4 Pro · high
63%
DeepSWE
Claude Code
GLM-5.2
44%
DeepSWE
Claude Code
Opus 4.8 · medium
49%
DeepSWE
Codex
GPT-5.5 · medium
54%
DeepSWE
AA Coding Agent Index · artificialanalysis.ai · captured 23.09.2026 · 6 pairs
DeepSWE v1.1 · deepswe.datacurve.ai · captured 22.09.2026 · 20 pairs
Terminal-Bench 4.0 · tbench.ai · captured 23.09.2026 · 13 pairs
RuBench round-01 · vibecoding.tech/rubench · 07/09/2026 · pass@1, xhigh