Notifications

Limited-time 50% off: Kling Video-to-Audio (kling-video-to-audio) drops from $0.044 to $0.022 per call until October 9, 08:37 UTC. It adds matching sound effects and background music to 3–20 second clips — same model name and request, discount applied automatically, charged only on success.

Retirement notice: grok-4.2-image will be retired on October 8, 2026, because our upstream cost now exceeds its price. The drop-in replacement is grok-imagine-image: the same model and request format at $0.03 per image, so you only change the model name. For a lower price, use FLUX.2 Klein 4B (flux-2-klein-4b, $0.006 per image) or GPT Image 2.5 Flare (gpt-image-2.5-flare, $0.008 per image at 1K with quality low).

A Cloudflare network incident in Asia-Pacific (multiple undersea cable failures) is slowing down or breaking some image uploads, so some tasks may fail or take longer. We are working on it, and failed tasks are not charged.

New models: Claude Opus 5.5 ($2.40 / $12.00, 40% below list), GPT-6 Sol ($1.00 / $5.00, half of list), GPT-6 Luna ($0.09 / $0.45, 10% below list) and TypeSafe's Jev decision model ($0.05 per 1M input tokens, output free). Same key, live now.

GPT Image 2.5 price cut, up to 57% lower: 2K high $0.06 → $0.04, 4K high $0.10 → $0.08, 2K max $0.35 → $0.15 per image. Every tier from high upward is cheaper, effective on all calls now.

Grok 4.5 and Grok 4.6 are 15% cheaper: Grok 4.5 drops from $1.5 to $1.275 per 1M input tokens and $5 to $4.25 output; Grok 4.6 from $1.765 to $1.50 input, $5.294 to $4.50 output, and $0.441 to $0.375 cached input. Both now sit 25%–36% below xAI list pricing, effective immediately. Same model names, endpoint and parameters, and you are still charged only on success.

Wan 3.0 standard tier is now 10% cheaper: 480P $0.05 → $0.045, 720P $0.10 → $0.09, 1080P $0.20 → $0.18 per second, effective immediately; the prime tier is unchanged. Same model name, endpoint and parameters

GPT Image 2.5 price cut, up to 33% off: 2K medium drops from $0.028 to $0.025 per image, 2K high from $0.09 to $0.06, and 4K high from $0.15 to $0.10. Applies to both gpt-image-2.5-flare and gpt-image-2.5-sunburst; all other tiers keep their current prices. Model, API and parameters are unchanged, and you are still charged only on success. Effective immediately.

GPT Image 2.5 now has per-quality pricing — from $0.008 per image. See the rates

GPT Image 2.5 is live: gpt-image-2.5-flare (default tier, about half the latency of GPT Image 2) and gpt-image-2.5-sunburst (highest fidelity). Billed per image on a 1K / 2K / 4K × low / medium / high grid from $0.005 (medium $0.011 / $0.023 / $0.022, high $0.045 / $0.09 / $0.085), about 15% below OpenAI list. Native transparent PNG, mask and reference-image editing, OpenAI SDK drop-in with one base_url change, no organization verification, failed requests are not charged.

Seedream 5.0 Pro price cut, about 60% off: 1K drops from $0.076 to $0.03 per image and 2K from $0.15 to $0.06. Model, API and parameters are unchanged (text-to-image, single-image editing, multi-image fusion with up to 10 references, single 1K/2K output); billed only on success. Effective immediately.

Following the GPT-6 launch, upstream capacity is tight and gpt-image-2 is currently taking about twice as long as usual to generate — longer at peak. Jobs still complete normally, and failed tasks are never billed. We are sorry for any disruption to your work.

Price cut for GPT-5.6 Luna and Claude Sonnet 5, effective immediately (per 1M tokens): gpt-5.6-luna input $0.221 → $0.16, output $1.324 → $0.96, cached input $0.022 → $0.016; claude-sonnet-5 input $2.353 → $1.60, output $11.765 → $8.00, cache read $0.147 → $0.10, cache write $2.059 → $1.40. Both now sit 20% below their official list prices (OpenAI lists Luna at $0.20 / $1.20; Anthropic lists Sonnet 5 at $2 / $10). Context: OpenAI cut the GPT-5.6 list prices on August 21 and we have followed. The models, endpoints and parameters are unchanged, you are still charged only on success, and the new rates apply automatically with no code change.

Grok Video 3 pricing and capability update: grok-video-3 is billed per second — $0.04/s at 480p and 720p (6s $0.24, 10s $0.40, 15s $0.60) — and now also supports 1080p output at $0.12/s (6s $0.72, 10s $1.20, 15s $1.80). grok-video-3-10s keeps its six flat per-clip tiers by resolution and length: 720p $0.30 / $0.50 / $0.70 and 1080p $0.35 / $0.65 / $2.00 for 6 / 10 / 15 seconds. The API and parameters are unchanged and you are still charged only on success; grok-imagine-video-1.5 is not affected.

Price update for Kling Video-to-Audio (kling-video-to-audio): $0.044 per call, matching Kling Sound Effects. The model page and docs already listed $0.04, but a configuration gap meant calls were billed at $0.003 — now corrected to $0.044. Capabilities and the API are unchanged (SFX + background music for 3-20s video, with ASMR mode), and you are still charged only on success.

Price update for GPT Image 2 Lite (gpt-image-2-lite): the 1K tier goes from $0.008 to $0.01 per image, following exchange-rate-driven increases in upstream cost. 2K ($0.015) and 4K ($0.025) are unchanged, and the API is unchanged.

Wan 3.0 Video is 20% off through September 10: 480p $0.0596 → $0.0477/s, 720p $0.1191 → $0.0953/s, 1080p $0.2382 → $0.1906/s, and the high-speed Prime tier is discounted too (from $0.0714/s). One model name covers text-to-video, first/last-frame control and four-modality omni-reference (10 images + 5 video clips + 5 audio clips, or a document / web page), with native single-shot output up to 30 seconds at 30fps and a synced audio track. The discount applies automatically — no parameter to change — and failed tasks are never billed.

Price update for Nano Banana Pro (gemini-3-pro-image-preview): 1K and 2K go from $0.06 to $0.068, and 4K goes from $0.12 to $0.13. Both tiers stay well below Google's official rate — 49% below at 1K/2K and 46% below at 4K. The model, endpoint and parameters are unchanged, you are still charged only on success with failures never billed, and results are kept for 7 days. No other model's price is affected. Effective immediately. (Correction: an earlier version of this notice said 4K was coming *down* to $0.068. 4K is $0.13; this notice supersedes it.)

Price update for Kling Sound Effects (kling-sound-effects): $0.03 → $0.044 per call, following an upstream cost increase. Capability and API are unchanged (text to 3-10s sound effects), and you are still only charged on success with failures refunded. All other Kling models keep their current prices.

Seedance 2.5 price cut 10%: 480p drops from $0.134 to $0.120/s and 720p from $0.300 to $0.270/s; video editing and extension move to $0.072/s (480p) and $0.162/s (720p). Same model, same endpoint, same parameters — still settled on real token usage, charged only on success and refunded on failure. Effective immediately.

GLM-5.2 price cut: cached input $0.4 → $0.28 per 1M tokens (−30%), input $1 → $0.9 and output $3.5 → $3.15 (−10% each). Long multi-turn contexts benefit most — over 90% of this model's input tokens are cache hits. The primary upstream has also been upgraded, with no change to reliability. Effective immediately.

Qwen Image 3.0 price cut: standard drops from $0.04 to $0.035 per image (same price at 1K and 2K); Pro goes from $0.05 to $0.037 at 1K and from $0.10 to $0.075 at 2K — about a quarter off. Reference images are still $0.004 each, the endpoint, quality and parameters are unchanged, and failures are never charged.

GPT Image 2 price cut: 2K drops from $0.04 to $0.03 per image and 4K from $0.06 to $0.05; 1K stays at $0.025. Applied to both gpt-image-2 and gpt-image-2-all, with the 4K low tier moving to $0.05 as well. Nothing else changes — same quality tiers, same 16-image reference limit, same endpoint, and failures are still never charged.

Dreamina Seedance 2.0 price cut: Mini $0.11 → $0.048/s (about -56%), Fast $0.18 → $0.1372/s (about -24%); the standard tier, including native 4K, is unchanged. Nothing was cut to get there. Note: a photograph of a real person now works as a plain image URL — the old asset:// form is no longer supported on this line, so pass the URL instead.

Grok Imagine Image channel upgrade (effective now): it has moved off the old channel, the faulty image-editing output is fixed, and the price is unchanged at $0.03 per image. grok-4.2-image is now merged into grok-imagine-image — the old model name keeps working at the same price with no code change, and responses echo grok-imagine-image as the model. Mask inpainting has been removed along with the old channel, as it was an endpoint feature rather than a model capability; use gpt-image-2 if you need masked region editing.

A temporary system service issue earlier caused some models to be briefly unavailable. Everything has now been fully restored. We sincerely apologize for any inconvenience.

GPT Image 2 Lite (gpt-image-2-lite) has been retired: its channel was unstable, so calls now run on the more stable GPT Image 2 All channel (1K + low quality, cheaper). Existing gpt-image-2-lite calls keep working with no changes; new integrations should use gpt-image-2-all.

GPT Image 2 pricing update: now billed by resolution — 1K $0.025, 2K $0.04, 4K $0.06 (previously a flat $0.03/image). Effective immediately.

GPT Image 2 price drop: the whole series (gpt-image-2 / Lite / Beta) is now a flat $0.03 per image, effective now.

Due to exchange-rate changes, Kling V2.1 text-to-image pricing has increased from $0.012 to $0.016 per image, effective immediately.

Grok Imagine Image price drop: now $0.03 per image (was $0.035), effective now.

GPT Image 2 price drop: now a flat $0.06/image (size- and count-agnostic), effective immediately.