Leaked Gemini 4 Pro specs: 2M context at ~1/5 rival pricing would reset frontier economics
Wanderspor · reddit · 2026-09-27
A Reddit post aggregates circulating leaks and Arena.ai stealth tests (under the gemini-3.8-flash placeholder) for Google's rumored Gemini 4 Pro (unverified). Claimed benchmark leads: DeepSWE v1.1 88.7%, Terminal-bench 2.1 95.3%, GDPval-AA v2 2064 Elo (only model over 2000), OSWorld-2.0 86.8% — pointing at long-horizon agentic work. The rumored pricing is the wilder part: $2.25 input / $11.25 output with 2M context, roughly 1/5 of GPT-6 Astra ($12/$60, 1M) and Claude Fable 5.1 ($10/$50). The author attributes the jumps to strategy-level improvement (à la Google's Dream-RSI paper) plus compute scaling rather than true RSI, and advises holding off on long-term API commitments until official model cards drop.
Related event: Suspected Gemini 4 Pro Spotted in Anonymous Arena Testing(8 posts)→
More from Models
- Claude 3 Sonnet spirals into surreal gibberish, then snaps back to refusals — a funny look at post-training personas — repligate · 2026-11-16
- AHA!: weight-space programming with no LLM lifts Qwen3.5-0.8B GSM8K from 14.0 to 71.2 — hbouammar · 2026-11-16
- Gemini subscriptions reportedly coming to Hermes, with Hydrogen teased — lyraxana · 2026-11-16
- FutureSim Q3 eval: Opus 5.5 dethrones GPT, spawns 50+ subagents in 90-day simulation — maksym_andr · 2026-11-16
- Percy Liang to AI labs: share non-vacuous info outsiders can actually evaluate — lateinteraction · 2026-11-16
- Anthropic model personality watch: Mythos 5 'inhabits the god frame,' 5.1 doesn't — repligate · 2026-11-16