GPT-6.1 Sol vs Sonnet 5.5: Same Price. Ten Times Apart. What's the Catch?

Оригинальный автор：Proof of AI

Первоисточник：https://www.youtube.com/embed/L3r4mZKWiTs

GPT-6.1 Sol and Claude Sonnet 5.5 cost exactly the same per token ($2 in, $10 out per million). Yet in Artificial Analysis's tests one task cost $0.72 on Sol and $7.60 on Sonnet, ten times apart. So what's the catch? We put both models through eight head-to-head builds (a voxel fighter jet, a chess board, a Boeing 747, a realistic face, a robot warehouse, a robot arm, a draggable curtain and an SVG pelican), then went through the independent boards, the catches (speed, a hidden price step above 272K tokens, OpenAI's own safety numbers) and one developer's real bills. Along the way: what OpenAI announced at DevDay 2026, and why GPT-6.1 Astra was scrapped a day earlier. Your call in the comments: which one is your default? A: Sol, the cheap engine. B: Sonnet, the top score. Next up: how Claude Opus 5.5 fits in: https://youtu.be/t-bS321Mh-0 CHAPTERS 0:00 Same price, ten times apart 0:31 The setup 1:19 The eight rounds 3:33 The final score 3:47 The independent boards 4:41 The three catches 5:40 Which one to pick SOURCES • OpenAI, Introducing GPT-6.1 Sol (29 Sept 2026): https://openai.com/index/introducing-gpt-6-1-sol/ • OpenAI DevDay 2026 recap: https://openai.com/index/devday-2026-recap/ • OpenAI DevDay 2026 keynote (full): https://www.youtube.com/watch?v=Fls_onRviPM • OpenAI, Ultrafast clip: https://www.youtube.com/watch?v=sKltfHvsDQM · dots clip: https://www.youtube.com/watch?v=uXspbC2srEQ · Codex in the cloud: https://www.youtube.com/watch?v=7Bv68f5szSU • OpenAI docs, GPT-6.1 Sol and pricing: https://developers.openai.com/api/docs/models/gpt-6.1-sol · https://developers.openai.com/api/docs/pricing • OpenAI safety report for GPT-6.1 Sol: https://deploymentsafety.openai.com/gpt-6-1-sol • GPT-6.1 Astra cancelled: https://www.engadget.com/2271626/openai-cancels-gpt-6-1-astra-release-deceptive-behavior/ · https://gizmodo.com/with-no-astra-to-release-openai-pivots-to-new-gpt-6-1-sol-model-2000819044 • Anthropic, Claude Sonnet 5.5: https://www.anthropic.com/claude-sonnet-5-5 • Artificial Analysis: https://artificialanalysis.ai/models/gpt-6-1-sol · https://artificialanalysis.ai/models/claude-sonnet-5-5 • Vals Index: https://www.vals.ai/benchmarks/vals_index • Theo (t3.gg), "Sol 6.1 shouldn't be so cheap": https://www.youtube.com/watch?v=vu8X3YroB-w • Vellum, GPT-6.1 Sol benchmarks explained: https://www.vellum.ai/blog/gpt-6-1-sol-benchmarks-explained CREATIONS SHOWN (credit to their authors) • Voxel fighter jet: MineBench sandbox, same prompt for both models, recorded by us: https://minebench.ai/leaderboard/gpt-6-1-sol · https://minebench.ai/leaderboard/sonnet-5-5 (MineBench lists Sol as "GPT 6.1 Sol Pro"; its ranking for Sol is still provisional) • Chess board, Boeing 747, realistic face, robot warehouse, robot arm, draggable curtain: threejseval.com, votes from its public gallery: https://threejseval.com/gallery • SVG pelican on a bicycle: Simon Willison, https://simonwillison.net/2026/Sep/29/hn-49898129/ · https://simonwillison.net/2026/Sep/28/claude-sonnet-5-5/ • DevDay footage: OpenAI. What is confirmed: prices, benchmark numbers, cost per task and OpenAI's own safety figures (vendor pages and the independent sources above). What is small-sample or our own call: the round verdicts (each three.js scene has only a few votes, and MineBench's ranking of Sol is provisional), the single pelican sample for Sonnet, and Theo's bills, which come from his own test harness. Music: Mixkit (mixkit.co). DevDay footage: OpenAI. Proof of AI · Proof before hype · @GetProofOfAI #GPT61 #ClaudeSonnet #AInews