Gonka community portal, created and maintained by community members

Kimi-K2.6 lost its majority of compute verification votes after two backup servers failed; the rest of the network continued operating.

LLM tokens this week
77.3B
DevShard · input + output
at least Pool trading volume
$175k
Uniswap v3 WGNK/USDT 0.3% on Ethereum
Reporting period
13—19
July 13–19, 2026, UTC
LLM tokens per day
LLM tokens per day08.2B16.4B13 Jul19 Jul
Current periodPrevious period
DevShard · input + output
GNK price · daily close, USD
GNK price · daily close, USD$0.128$0.150$0.17313 Jul19 JulIncomplete data · 166/168

GNK trading volume reached at least $175.6 thousand

Measured GNK trading volume totaled at least $175.6 thousand. The price at the start of the period was $0.1734. No closing price is available, so no weekly price change or volume comparison with the previous week is reported.

Completed requests rose by 40.8%

DevShard statistics show 7.88 million completed requests, compared with 5.60 million for July 6–12. It processed 66.18 billion input tokens and 11.07 billion output tokens. Total token volume rose by 30.2%. MiniMax-M2.7 led with a 77.7% share of tokens, while Kimi-K2.6 accounted for 22.3%.

Kimi removal and restoration approved

Proposal #86 to increase the weight coefficients for Kimi-K2.6 and GLM-5.2 by 5% was approved on July 16. Proposal #87 to remove Kimi from the compute verification parameters and model list was approved the same day. On July 17, proposal #88 was approved to return it to the list and exclude older DevShard versions. Removal and restoration were recovery steps using the standard model launch process.

Stability, penalties and pricing

Participants discussed Kimi and MiniMax timeouts and rate-limit errors. Experiences varied: Kimi worked normally for some users. In a related debate, some participants proposed removing penalties for not delegating models and switching to positive incentives. Views on pricing differed: some suggested accounting for hosts’ actual costs, while others considered an increase premature until quality improves.

Older DevShard processes stopped

Documentation dated July 17 confirms that DevShard v1/v2 processes have been stopped. Gateway traffic must go through v3; gateways that stopped serving inference because of old routes need to migrate. This does not confirm that all gateways have completed the transition. Following the incident, the recommendation to delegate to backup nodes was also withdrawn, and spreading delegations across independent hosts was suggested.

← All editions