OpenCode makes DeepSeek V4.1 Flash free, capped at 200K context
OpenCode announced that DeepSeek V4.1 Flash is now offered free on its platform, with the free tier observed to support a 200K context window, down from the model's 1M-context version. The move extends the earlier "Operation Cheepseek" program: OpenCode temporarily raised the model's allowance in its OpenCode Go plan from $15 to $60 in September, then to $80 on October 12, and calls under the free offer no longer count against that allowance. DeepSeek V4.1 Flash launched on September 10, 2026 as a native multimodal MoE model with a 552B-parameter main trunk plus 196B Engram conditional-memory parameters, a causal encoder-decoder design in which the first 20 of 40 language layers are encoders and the last 20 decoders, cutting prefill compute roughly in half; DeepSeek's own platform lists no free API tier. Community reports flag volatile throughput: one user saw up to about 490 tokens/s on the ollama cloud platform, but speeds can fall to 50-60 tokens/s around midday to afternoon and decline as context grows, reaching roughly 50 tokens/s at 200K.
Free access lowers the barrier to trying the model, but the reduced 200K context and unstable throughput suggest the offer works more as a funnel than as a like-for-like alternative to paid access.