GPT 6.2 Sol
An AI model released by OpenAI, with demonstrated performance in areas such as cybersecurity.
Core Facts
Timeline (last 90 days)
Google DeepMind released the Gemini 4 Neon model, targeting Claude Opus 5.5, GPT 6.2 Sol, and Fable 5.2.
On Frontier code 2.0, Neon was about one percentage point lower than Opus but higher than GPT 6.2 Sol.
On the AA index, Neon's score exceeded those of Opus 5.5 and GPT 6.2 Sol.
On Agent Arena, covering 4,000 real user sessions, Neon at the default high gear ranks third, at $2.60 per session, about half of Opus's $5.89 and lower than GPT 6.2 Astra's $3.18; GPT 6.2 Sol remains at the top.
In a one-shot cathedral visual generation test, Neon cost $2.67 and GPT 6.2 Sol cost $1.69; Neon's output included flying buttresses, arched stained-glass windows and a vaulted interior hall with red carpet.
Opus 5.5 leads by about 1 point on Terminal Bench 4.1 and Frontier Code 2.0, while Sol 6.2 has a slight edge on Chartography.
Neon ranks first globally on the Artificial Analysis Coding Agent Index with 58.6 points and first on Zapier's Automation Bench with 56.3%, leading Opus 5.5 by more than 13 points and Sol 6.2 by 15 points respectively.
In Artificial Analysis's Omniscience test, Neon's non-hallucination rate was 85.6%, rising to 86.1% at the default high thinking setting, nearly double Claude Opus 5.5's 49.7%, while GPT 6.2 Sol was 44.6%.
In GraySwan's indirect prompt injection test, Neon's attack success rate was 0.0%, Opus 5.5's was 1.0%, and Sol 6.2's was 2.3%.
At the ultra-high thinking setting, Neon cost $6.44 per task to complete the full benchmark suite, 36% cheaper than Opus 5.5 Max's $10.04, and its time to first answer was six times faster (56.5 seconds vs 338.6 seconds); GPT 6.2 Sol cost $2.10 per task, but Neon's output speed of 110 tokens per second was 47% faster.
9 more timeline events
All Facts (18)
At the ultra-high thinking setting, Neon cost $6.44 per task to complete the full benchmark suite, 36% cheaper than Opus 5.5 Max's $10.04, and its time to first answer was six times faster (56.5 seconds vs 338.6 seconds); GPT 6.2 Sol cost $2.10 per task, but Neon's output speed of 110 tokens per second was 47% faster.
60%UnverifiedA Bilibili creator ran a three-round hands-on test comparing the real-world performance of Gemini 4 Neon, Opus 5.5 and GPT-6.2 Sol.
60%Partially VerifiedOpenAI's GPT-6.2 series includes two separate models: Astra and Sol, both launched on AI Gateway.
65%UnverifiedOn the AA index, Neon's score exceeded those of Opus 5.5 and GPT 6.2 Sol.
50%UnverifiedOn Frontier code 2.0, Neon was about one percentage point lower than Opus but higher than GPT 6.2 Sol.
50%UnverifiedIn a one-shot cathedral visual generation test, Neon cost $2.67 and GPT 6.2 Sol cost $1.69; Neon's output included flying buttresses, arched stained-glass windows and a vaulted interior hall with red carpet.
50%UnverifiedOn Agent Arena, covering 4,000 real user sessions, Neon at the default high gear ranks third, at $2.60 per session, about half of Opus's $5.89 and lower than GPT 6.2 Astra's $3.18; GPT 6.2 Sol remains at the top.
50%UnverifiedNeon ranks first globally on the Artificial Analysis Coding Agent Index with 58.6 points and first on Zapier's Automation Bench with 56.3%, leading Opus 5.5 by more than 13 points and Sol 6.2 by 15 points respectively.
50%UnverifiedOpus 5.5 leads by about 1 point on Terminal Bench 4.1 and Frontier Code 2.0, while Sol 6.2 has a slight edge on Chartography.
50%UnverifiedIn Artificial Analysis's Omniscience test, Neon's non-hallucination rate was 85.6%, rising to 86.1% at the default high thinking setting, nearly double Claude Opus 5.5's 49.7%, while GPT 6.2 Sol was 44.6%.
50%UnverifiedIn GraySwan's indirect prompt injection test, Neon's attack success rate was 0.0%, Opus 5.5's was 1.0%, and Sol 6.2's was 2.3%.
50%UnverifiedOn the Artificial Analysis index, Neon surpassed Gemini 4 Carbon and also surpassed the best Opus and GPT 6.2 Sol.
50%UnverifiedOn the Cursor Bench 4.1 test, Gemini 4 Neon scored 65.4% and Claude Opus 5.5 scored 58.1%.
50%UnverifiedIn the long-context test GraphWalks, Neon scored 95.2% while Opus scored only 66.8%.
50%UnverifiedIn the code test SWE Sweep, Neon scored 39% versus 10.1% for Opus; but when a common tool is used uniformly, Neon scores 12.8%, close to Sol's 11.6%, indicating that the 39% result benefits from the advantage of Google's own Anti-Gravity tool.
50%UnverifiedGoogle's own comparison table pits Neon against Claude Opus 5.5 and GPT 6.2 Sol; Neon won 16 of 19 tests, and the three it lost were the terminal task Terminal Bench and the code task Frontier Code (both lost to Opus) and professional chart reading (lost to Sol), all by margins within two percentage points.
50%UnverifiedAn evaluation by an LMArena creator lasting 44 minutes and covering nearly a hundred scenes concluded that the Sol version stands out on value, costing about 1/5 to 1/10 of the comparison model.
50%UnverifiedGoogle DeepMind released the Gemini 4 Neon model, targeting Claude Opus 5.5, GPT 6.2 Sol, and Fable 5.2.
50%