Opus 5.5 beats GPT-6 Sol 6-1 in seven-task showdown

Anthropic's Opus 5.5 and OpenAI's GPT-6 Sol launched minutes apart on the same day, and blogger @alexverem immediately scored them across seven representative tasks. Opus 5.5 won 6-1, losing only the lava lamp round where Sol's speed and cost prevailed. The benchmark covers creative generation, engineering quality, hardcore GPU work, and subscription value, making it a widely cited reference for comparing the two flagships.

Confirmed

• Round 1 (Three.js Eiffel Tower, four models): Opus 5.5 cost $8.95/10 min, GPT-6 Astra $7.45/8 min, GPT-6 Sol $3.90/5 min, with Kimi K3 also participating; the author found Opus's result the most impressive.
• Round 2 (lava lamp): GPT-6 Sol won in 50 seconds at 8 cents; Opus 5.5 took 9 min 35 s at $1.27; GPT-6 Astra $0.63/5 min 4 s; GPT-6 Luna under 1 cent.
• Round 3 (Blender scene, same prompt): Sol cost $34.55 and was faster but had noticeable rigging and geometry problems; Opus spent $48.11, was slower, but produced far more polished detail and atmosphere.
• Round 4 (single-HTML fishing game): Sol was faster with smaller, cleaner code but had only 1 fish species and 2 upgrades; Opus built 5 fish species, rarity tiers, a perfect-catch mechanic, 3 skill trees, dynamic schools, a minimap, and more.
• Round 5 (same-prompt games): the author called the gap "not even close" and said Opus 5.5 is the best model he has seen; the round went to Opus.
• Round 6 (Intel Arc B70): Opus 5.5 got exl3 running in 2 hours and fixed decode, prefill, concurrency, MTP, and vision; the author said weeks of attempts with Sol had failed.
• Round 7 (subscription value): on Claude Pro ($20), Opus 5.5 high ran nearly 2 hours before hitting the 5-hour limit, with each window only 15% of weekly usage; the author deemed this far above what $20 ChatGPT Plus offers.
• Final score: Opus 6 wins (Eiffel Tower, Blender, fishing game, same-prompt game, GPU task, subscription quota), Sol 1 win (lava lamp).

Unconfirmed

• All conclusions come from @alexverem's solo benchmark; he self-identifies as preferring Opus, so results carry subjective judgment and have not been independently replicated.

Why it matters

• Same-day launches of two flagship models are rare, and this seven-round duel spans creative generation, engineering quality, and cost efficiency, offering a multi-dimensional selection reference. In particular, the long-horizon Intel Arc B70 task and the $20 subscription quota test bear directly on heavy users' real costs and productivity judgments.

2026-09-25 ~ 2026-09-25 · 9 related posts

• [source] Opus 5.5 and GPT-6 Sol launched minutes apart; author scores 7 matchups — alex_verem · 2026-09-25
• Three.js Eiffel Tower, four models: Opus 5.5 shines, Kimi K3 best value — alex_verem · 2026-09-25
• Lava lamp test: GPT-6 Sol wins on cost and speed at $0.08 in 50 seconds — alex_verem · 2026-09-25
• Blender head-to-head: Sol saves ~$14, Opus delivers the polished scene — alex_verem · 2026-09-25
• Fishing game test: Sol optimizes the benchmark, Opus builds the product — alex_verem · 2026-09-25
• Same-prompt game build: Opus 5.5 vs GPT-6 Sol "not even close" — alex_verem · 2026-09-25
• Opus 5.5 gets exl3 running on Intel Arc B70 GPUs in 2 hours — alex_verem · 2026-09-25
• $20 Claude Pro runs Opus 5.5 high for nearly 2 hours per 5-hour window — alex_verem · 2026-09-25
• [source] Final tally: Opus 5.5 dominates GPT-6 Sol 6-1 across seven tests — alex_verem · 2026-09-25