Thinking Machines' ReViSQL-K2.6 hits human-level text-to-SQL accuracy
Start here
Thinking Machines has detailed ReViSQL-K2.6, a Kimi-K2.6 model fine-tuned with reinforcement learning on a cleaned training set, which reaches 92.97% on Arcwise-Plat-SQL using 16-sample self-consistency at $0.56 per task — just above the 92.96% human proxy. The team says the model beats GPT-5.6 Sol Ultra and Claude Fable 5 at 12–15% of their cost. The result rests on a corrected benchmark: the team audited 2,500 BIRD Train instances and found errors in 61.1% of them, including incorrect gold SQL in 52.1%, and released the fixed set as BIRD-Platinum. Reward shaping adds VeriEQL equivalence checks and rule-based process rewards for external knowledge. The accuracy and cost comparisons are the team's own reported figures.
Generated from reports · updated 6 h ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AIThinking Machines details ReViSQL-K2.6, an RL-trained text-to-SQL model reaching human-level accuracy without scaffolding
Thinking Machines describes ReViSQL-K2.6, a Kimi-K2.6 model fine-tuned with RLVR on a cleaned training set, which reaches 92.97% on Arcwise-Plat-SQL with 16-sample self-consistency at $0.56 per task, above the 92.96% human proxy. The team audited 2.5k BIRD Train instances and found errors in 61.1% of them, including incorrect gold SQL in 52.1%, then released the corrected set as BIRD-Platinum. Reward shaping adds VeriEQL equivalence checks and rule-based process rewards for external knowledge, and the model beats GPT-5.6 Sol Ultra and Claude Fable 5 at 12–15% of their cost.
Heat for this event
- Current, comparable range
- 9
- Peak, comparable range
- 10November 16 11:00
- Change in 24 hours
- –
The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.