Open-weight models are now ~90% as good as frontier models at 10–25% of the token cost, pushing AI on-prem
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 26 October 2026
Your AI and this story
- GPT-6.2 Astra144 days after its cutoff
- Claude Opus 5.5114 days after its cutoff
- Gemini 4 Hydrogen205 days after its cutoff
- Grok 4.7144 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 114 days before it.
Key facts
- Krishna: open-weight models cost ‘anywhere from 25 percent to 10 percent of the token cost of the frontier models’; ‘Would you like something that’s 10 percent of the cost if it’s 90 percent as good as the best’
- His illustration of the gap: in 2024 ‘the top model might score a 70, and the open model might score a 20 or a 30’; ‘In late 2025, we began to see that gap shrink’ (to about 70 vs 50). He named no models or benchmark
- Claim: most enterprise work (summarising email, reviewing contracts) does not need top models; most businesses could run useful on-prem AI in two or three racks; IBM expects this to lift sales of its Spyre accelerator cards for mainframes
- Quarter (per The Register): revenue $17.1bn (+5%); software $7.8bn (+9%), Red Hat +11%; infrastructure $3.7bn (+5%)
- Reach: 411 points on Hacker News (Oct 26)
What happened
On its quarterly call IBM’s chief executive argued that open-weight models have become good and cheap enough that enterprises will run much of their AI on their own hardware. He gave a stylised scoring example rather than a named benchmark.
Why it matters
A large enterprise vendor publicly betting on open weights is one side of the same price pressure that drove OpenAI’s and Anthropic’s 2026 API price cuts.
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 1 press, 1 reaction
Press
- The Register: IBM CEO says open weight models are now irresistibly good and cheap on-prem (Oct 22)theregister.com, press
Reactions
- Hacker News discussion (411 points)news.ycombinator.com, discussion
Changes
- Filed (Hacker News)