Claude Opus 4.8 holds #1 on June leaderboards: 61.4 Artificial Analysis Intelligence Index, tops Appwrite Arena at 97.4%
Across early-June 2026 evaluation updates, Anthropic's Claude Opus 4.8 — released May 28 at unchanged pricing of $5/$25 per million input/output tokens with a 1M-token context window — consolidated its position as the to.
At a glance
- Claude Opus 4.8 released May 28; pricing unchanged at $5/$25 per million input/output tokens; 1M-token context window
- Appwrite Arena June update (added four new frontier models): #1 without-skills at 97.4%, first model to beat Opus 4.7
- Artificial Analysis Intelligence Index 61.4: +4.1 over Opus 4.7, +1.2 over prior leader GPT-5.5 (xhigh)
- GDPval-AA 1,890 Elo: +137 over Opus 4.7, +121 over GPT-5.5 (xhigh), implying ~67% win rate
- Terminal-Bench Hard +6.8; τ²-Bench Telecom +5.9; IFBench +3.6; SWE-bench Pro 69.2% (from 64.3%); Online-Mind2Web computer use 84%
VERDICT — CONFIRMED
Across early-June 2026 evaluation updates, Anthropic's Claude Opus 4.8 — released May 28 at unchanged pricing of $5/$25 per million input/output tokens with a 1M-token context window — consolidated its position as the top-ranked frontier model. In Appwrite Arena's June update, which added four new frontier models, Opus 4.8 took #1 on the 'without-skills' board at 97.4%, the first model to beat its predecessor Opus 4.7. On the Artificial Analysis Intelligence Index it scored 61.4, a +4.1 jump over Opus 4.7 and +1.2 ahead of the prior leader GPT-5.5 (xhigh).
The headline agentic result: 1,890 Elo on GDPval-AA (Anthropic's primary knowledge-work agent benchmark), +137 over Opus 4.7 and +121 ahead of GPT-5.5 (xhigh), implying a ~67% win rate against GPT-5.5. Other gains include Terminal-Bench Hard (+6.8 points), τ²-Bench Telecom (+5.9), IFBench (+3.6), SWE-bench Pro 69.2% (up from 64.3%) and Online-Mind2Web computer use at 84%. On the AA-Omniscience knowledge eval it placed second at 27.4 (46.6% accuracy, 35.9% hallucination rate).
BenchLM ranked it #2 of 119 models overall (95/100) and #2 in coding (98.9 average). Anthropic also emphasized alignment: Opus 4.8 is reportedly four times less likely to overlook code flaws and set 'new highs on prosocial traits' with lower misaligned-behavior rates than Opus 4.7. The cluster confirms a tightening top tier where Anthropic narrowly leads on agentic and coding evals while saturation on MMLU-class benchmarks pushes attention to GPQA, GDPval and agentic suites.
Key facts on file
- Claude Opus 4.8 released May 28; pricing unchanged at $5/$25 per million input/output tokens; 1M-token context window
- Appwrite Arena June update (added four new frontier models): #1 without-skills at 97.4%, first model to beat Opus 4.7
- Artificial Analysis Intelligence Index 61.4: +4.1 over Opus 4.7, +1.2 over prior leader GPT-5.5 (xhigh)
- GDPval-AA 1,890 Elo: +137 over Opus 4.7, +121 over GPT-5.5 (xhigh), implying ~67% win rate
- Terminal-Bench Hard +6.8; τ²-Bench Telecom +5.9; IFBench +3.6; SWE-bench Pro 69.2% (from 64.3%); Online-Mind2Web computer use 84%
- AA-Omniscience second at 27.4 (46.6% accuracy, 35.9% hallucination rate); BenchLM #2 of 119 (95/100), #2 coding (98.9 average)
- Opus 4.8 reportedly four times less likely to overlook code flaws; 'new highs on prosocial traits'
